Computational molecular phenotyping of retinal sheet transplants to rats with retinal degeneration

Abstract

Retinal progenitor sheet transplants have been shown to extend neuronal processes into a degenerating host retina and to restore visual responses in the brain. The aim of this study was the first attempt to identify cells involved in transplant signals to retinal degenerate hosts using computational molecular phenotyping (CMP). - S334ter line 3 rats received fetal retinal sheet transplants at the age of 24-40d. Donor tissues were incubated with slow-releasing microspheres containing BDNF or GDNF. Up to 265 days after surgery, eyes of selected rats were vibratome sectioned through the transplant area (some slices stained for donor marker hPAP), dehydrated and embedded in Eponate, sectioned into serial ultrathin datasets and probed for rhodopsin, cone opsin, CRALBP, L-glutamate, L-glutamine, glutathione, glycine, taurine, GABA, and DAPI. - In large transplant areas, photoreceptor outer segments in contact with host RPE revealed rod and cone opsin immunoreactivity whereas no such staining was found in the degenerate host retina. Transplant photoreceptor layers contained high taurine levels. Glutamate levels in the transplants were higher than in the host retina whereas GABA levels were similar. The transplant inner nuclear layer showed some loss of neurons, but amacrine cells and horizontal cells were not reduced. In many areas, glial hypertrophy between the host and transplant was absent and host and transplant neuropil appeared to intermingle.

CMP data indicate that horizontal cells and both glycinergic and GABAergic amacrine cells are involved in a novel circuit between transplant and host, generating alternative signal pathways between transplant and degenerating host retina.

Introduction

Retinal degenerative diseases such as age-related macular degeneration (AMD) or Retinitis Pigmentosa (RP) lead to progressive vision loss and, ultimately total blindness in millions of people worldwide. When photoreceptor loss occurs, the neural retina underneath demonstrates plasticity, and continues to signal in the absence of photoreceptor input (Margolis et al., 2008; Jensen & Rizzo, 2009; Kolomiets et al., 2010), albeit with early reprogramming in glutamate receptor expression and alterations in retinal circuitry (Strettoi et al., 2003; Marc et al., 2007; Marc et al., 2008; Lin et al., 2009). This prolonged function along with demonstrated plasticity in the remnant neural degenerate retina may define windows of opportunity for intervention (Jones et al., 2003) and raises the possibility of therapies based upon the replacement of diseased cells with retinal progenitors. Transplanted cells may be able to integrate into the remodeling retina and form functional connections with the host retina, ultimately assisting with retinal repair and vision improvement.

Methodologies have been developed to transplant freshly harvested sheets of fetal retinal progenitor cells with or without retinal pigment epithelium to the subretinal space (Aramant & Seiler, 2004; Radtke et al., 2008; Seiler et al., 2008a) which preserve the three-dimensional topology, resembling a normal retina (review: Aramant & Seiler, 2004). In contrast, with dissociated retinal progenitor cells, photoreceptors do not survive in long-term transplants (Mansergh et al., 2010) although - in microaggregate transplants - photoreceptors can survive for 8 months (Gouras & Tanabe, 2003).

Retinal sheet transplants restore visual responses in retinal degenerative models in a retinotopic manner in the superior colliculus (SC) (Woch et al., 2001; Sagdullaev et al., 2003; Thomas et al., 2004; Yang et al., 2010). SC recording is an established method to demonstrate photoreceptor rescue by retinal pigment epithelium (RPE) transplants (Sauvé et al., 1998). Synaptic connections between transplant and host are implied through trans-synaptic tracing from the visually responsive site in the SC back to the transplant with attenuated pseudorabies virus (Seiler et al., 2008c), and established directly by demonstrating synapses of transplant neurons in the host inner plexiform layer at the ultrastructural level with transplants that had restored visual responses in the SC (Seiler et al., 2010).

The retinal degeneration model used in this study was the transgenic pigmented S334ter line 3 rat (Steinberg et al., 1996) expressing a mutated truncated rhodopsin gene resulting in a protein lacking the 15 carboxyl-terminal amino acids involved in the rhodopsin transport to the outer segments (Lee & Flannery, 2007; Lin & LaVail, 2010). Similar mutations within the rhodopsin C terminus have been found in RP patients (Berson et al., 2002) and lead to photoreceptor cell death (Liu et al., 1999; Green et al., 2000).

The goal of the current study was to investigate the potential circuitry of long-term retinal progenitor sheet transplants that had restored responses in the SC, using computational molecular phenotyping (CMP), a technique to transform anatomical data into multidimensional signatures, resulting in cell classification and quantitative visualization as tissue theme maps (Marc & Jones, 2002; Jones et al., 2003).

Materials and Methods

Ethics statement

For all experimental procedures, animals were treated in accordance with the NIH guidelines for the care and use of laboratory animals and the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research, and under protocol 2006-2698 approved by the Institutional Animal Care and Use Committee of the University of California, Irvine (animal welfare assurance number A3416-01). All efforts were made to minimize animal suffering and to use only the minimum number of animals necessary to provide an adequate sample size.

Experimental animals

Donor rats with normal retina were bred from transgenic rats positive for human placental alkaline phosphatase (hPAP) (Kisseberth et al., 1999; Mujtaba et al., 2002); breeders were a gift of Dr. Eric Sandgren, Univ. of Wisconsin, Madison, WI) and ACI rats were obtained from Harlan Laboratories (Indianapolis, IN, USA). RD rats were the F1 generation of a cross between albino homozygous S334ter line 3 and pigmented Copenhagen rats (Harlan Laboratories). This line was originally produced by Xenogen Biosciences (formerly Chrysalis DNX Transgenic Sciences, Princeton, NJ), and supplied with the support of the National Eye Institute by Dr. Matthew LaVail, University of California, San Francisco (http://www.ucsfeye.net/mlavailRDrat models.shtml).

Donor tissue

Fetal donor retina was dissected from fetuses of timed-pregnant hPAP donors that were determined to be hPAP-positive transgenic by a histochemical reaction for hPAP (Kisseberth et al., 1999; Yang et al., 2010). Fetuses and dissected retinas were stored in Hibernate E medium on ice or at 4°C as previously described (Seiler et al., 2008b; Seiler et al., 2010). In some of the experiments, retinal flatmounts were incubated with BDNF or GDNF microspheres overnight (Seiler et al., 2008b; Yang et al., 2010). Before transplantation, retinas were cut to rectangular pieces of ca. 1 mm² size to fit the flat nozzle of a custom-made implantation tool.

Transplantation

Thirteen transgenic S334ter-line-3 RD rats expressing mutated human rhodopsin protein (Sagdullaev et al., 2003; Ray et al., 2010) were selected for this study (Table 1). Additional control rats for electrophysiology received transplants of fetal cortex or sham surgery. After ketamine/xylazine anesthesia (37.5 mg/kg ketamine; 5 mg/kg xylazine), rats received fetal retinal sheet transplants in one eye at the age of P24 to 41 (10 different surgery days), as previously described (Seiler & Aramant, 1998; Aramant & Seiler, 2002). Briefly, after pupil dilation with 1% atropine (Bausch & Lomb, Tampa, FL, USA), and local anesthesia by tetracaine eye drops (0.5%) (Bausch & Lomb), the sclera was exposed, and a small cut (~1.0 mm) was made parallel to the limbus and posterior to the pars plana. The implantation instrument was inserted into the subretinal space and advanced to the nasal quadrant close to the optic disk. Release of the instrument placed the tissue into the subretinal space, after which the instrument was withdrawn. The lesion was closed with 10-0 sutures and the eyes were treated with gentamycin and artificial tears ointment. Rats were placed in an incubator for recovery and correct placement of the transplant into the subretinal space was confirmed by postsurgical fundus exams and ocular coherence tomography.

Table 1. Overview of experiments.

Rat #	Sex	Age at surgery	Donor tissue	Survival time	Age at sacrifice	Remarks
1	F	40 d	E19 retina	256 d	296 d	Laminated area (7 of 12 slices) no recording
2	M	41 d	E19 retina + BDNF	249 d	290 d	Laminated area (6 of 12 slices) no recording
3	F	35d	E19 retina + BDNF	251 d	286 d	Laminated area (3 of 11 slices) no recording
4	F	24d	E19 retina + GDNF	200 d	224 d	Mostly rosettes, laminated area in center of graft (4 of 9 slices) no recording
5	F	32d	E19 retina + GDNF	87 d	129 d	Mostly rosettes, small laminated area (5 of 11 slices); SC recording, threshold −2.66 log cd/m2
6	F	35d	E19 retina + BDNF	251 d	286 d	Laminated area (7 of 11 slices) no recording
7	F	26 d	E19 retina + GDNF	130 d	155 d	Laminated area (9 of 13 slices) SC recording, threshold −2.2 log cd/m2
8	F	32 d	E19 retina + GDNF	130 d	162 d	Laminated area (5 of 12 slices) SC recording, threshold −1.5 log cd/m2
9	F	32 d	E19 retina + GDNF	77 d	110 d	Laminated areas (8 of 11 slices) SC recording, threshold −2.0 log cd/m2
10	M	33 d	E19 retina	88 d	121 d	Mostly rosettes, small laminated area (3 of 10 slices) SC recording, threshold −2.5 log cd/m2
11	F	37 d	E19 retina + BDNF	92 d	129 d	Mainly rosettes, tiny laminated area (5 of 12 slices) SC recording, threshold□−2.2 log cd/m2
12	F	25 d	E19 retina + GDNF	125 d	150 d	Rosettes only (7 slices) SC recording, threshold −0.9 log cd/m2
13	M	33 d	E19 retina + GDNF	69 d	102 d	Rosettes only (11 slices) SC recording, threshold −1.5 log cd/m2

Superior colliculus (SC) recording

The visual responses to a full-field light stimulus were recorded in 8 rats, 69-130 days after transplantation according to a previously described procedure (Thomas et al., 2005; Yang et al., 2010). Briefly, rats were dark-adapted overnight, anesthetized with Ketamine/Xylazine, and placed into a stereotaxic apparatus under isoflurane anesthesia. All procedures were performed in the dark under dim red light. After a craniotomy, the surface of the SC was exposed. Using tungsten microelectrodes (0.5 megaohms (MΩ) impedance; MicroProbe, Inc., Gaithersburg, MD), multi-unit extracellular responses were recorded systematically from various sites of the SC (at max. 0.3 mm depth) following a 40 ms light stimulus (luminance −5.0 to +1.0 log cd/m²), using a digital data acquisition system (Powerlab; ADI Instruments, Mountain View, CA). Stimuli were repeated at each site 10 times at 10 sec intervals. A response was defined across the set of 10 photic stimuli as >10% increase of spike counts over baseline or 10 spikes in the case of zero baseline activity. Visual responses were characterized for the location within the SC, response threshold, spike counts, and response latencies. A one-way analysis of variance (ANOVA) with subsequent post hoc tests from a statistics package of GraphPad Software, Inc. (San Diego, CA) was used for statistical comparisons (significance criterion P < 0.05).

Tissue processing

After anesthesia with an overdose of ketamine/xylazine, rats were perfused through the ascending aorta with 100 mL saline, followed by a 10-minute perfusion of 300 mL fixative (1% paraformaldehyde, 2.5% glutaraldehyde, 3% sucrose w/v, 1 mM MgSO₄ in 0.1 M sodium phosphate buffer pH 7.4). After enucleation and removal of the cornea, eyes were postfixed in the same fixative overnight, and washed 3 times with phosphate buffer. Eye cups were dissected to obtain cross-sections through the transplant area along the dorso-ventral (superior-inferior) axis, embedded in 4% agarose, and cut on a Leica VT 1000 S vibratome (Leica Instruments, Nussloch, Germany) at 70-100 μm. Some slices were stained for the donor label hPAP using BCIP/NBT (5-Bromo-4-chloro-3-indolyl phosphate / Nitro blue tetrazolium chloride; Sigma, St. Louis, B-1911) resulting in a blue-purple stain (Kisseberth et al., 1999; Yang et al., 2010). Selected vibratome slices were processed as previously described (Marc & Jones, 2002). Briefly, sections were dehydrated in graded methanols, followed by 100% acetone and infiltration with increasing concentrations of Eponate 812 (Ted Pella, Redding, CA) in acetone. After infiltration with 100% eponate, the slices were flat-embedded, sandwiched between 2 Aclar films (Electron Microscopy Sciences, Hatfield, PA) on glass slides, cured overnight at 65°C, cut out and glued onto Epon stubs for ultra-thin sectioning.

Electron microscopy

Semithin and ultrathin sections (70nm thick) were cut from selected BCIP-stained blocks. Ultrathin sections were either left unstained or stained for 10 sec. with uranlyacetate, and viewed in a JEOL1400 electron microscope at magnifications between 1,500 – 15,000x.

Quantification of donor-specific hPAP label in thin sections of transplanted fetal retinas in S334ter rat retinas

Serial 200 nm thin sections were imaged with as 8-bit 1388 pixel × 1036 line frames under voltage regulated tungsten halogen flux with a variation of 1.2 ± 0.6%/min (mean ± sd). Image mosaic tiles were captured with a Peltier-cooled QImaging Fast 1394 QICAM (QImaging, Burnaby, BC, Canada) and a Syncroscan montaging system (Synoptics Inc, Frederick, MD) on a Scan 100×100 stage (Märzhäuser Wetzlar GMBH & Co, Germany).

Every retina was imaged in a spectral series at 400, 450, 500, 550, 600, 650 nm to verify that the absorbance curve of the product matched the blue diformazan of NBT (Marc & Sperling, 1976). The tissue curve was very similar to aqueous suspensions of NBT diformazan but with a red shift characteristic of diformazans in non-aqueous media (Figure 2).

Figure 2. Quantification of transplant-specific NBT staining.

Absorbance of NBT reaction product versus wavelength. Every retina was imaged in a spectral series at 400, 450, 500, 550, 600, 650 nm to verify that the absorbance curve of the product matched the blue diformazan of NBT (Marc & Sperling, 1976). Then the tissue curve was very similar to aqueous suspensions of NBT diformazan but with a red shift characteristic of diformazans in non-aqueous media. All the transplanted regions showed characteristic NBT diformazan densities ranging from 0.06 to 0.1 at 550 nm. This corresponds to an NBT diformazan concentration of ≈700 mM (using standard diformazan extinction coefficients and a section thickness of 200 nm), which is characteristic of diformazan reactions run to saturation. None of the host peripheral retinas from any of the samples showed any measurable density at any wavelength. They were absolutely transparent. Thus the spectral absorbance data uniquely represent the presence of enzymatically reduced NBT. There is no intrinsic production of the diformazan in the S344ter host retina. Diformazan is insoluble in aqueous media, ethanol, acetone and non-diffusible. Thus the reaction product in the inner plexiform layer of S344ter host retina represents cellular elements from the transplanted retina entering the host. No diformazan is present in any nucleus, despite the immensely high concentration of product.

Immunohistochemistry

Tissue blocks of 6 rats were selected for analysis. Two rats (rats # 5 and 7, post surgery times 87 and 130 days) had undergone SC recording (see Table 1). The other 4 rats (200 – 256 days post surgery) did not undergo SC recording.

Blocks were sectioned into serial ultrathin datasets and probed with IgGs targeting, L-glutamate, L-glutamine, glutathione, glycine, taurine, GABA (Signature Immunologics, Inc., Salt Lake City, UT), rhodopsin rho1D-4 (a kind gift from Dr. Robert Molday, Univ. of British Columbia, Canada), LWS1 (red and green) cone opsin from Chemicon International Inc. (Temecula, CA), and CRALBP (cellular retinaldehyde binding protein; a generous gift from Dr. Jack Saari, Univ. of Washington, Seattle WA); and DAPI (4′6-diamidino-2-phenylindole dihydrochloride) (procedure after Marc & Jones, 2002).

Computational Molecular Phenotyping/Data Analysis

Computational molecular phenotyping (CMP) has been extensively described in the literature (Marc et al., 1995; Marc & Cameron, 2002; Marc & Jones, 2002). CMP fuses molecular and computational approaches to simultaneously visualize co-localized concentrations of small molecular species while preserving anatomical context at cellular resolution. Detection is accomplished with IgGs generated against epitopes of interest and the combination of signals defines individual cell classes. The immunocytochemistry is transformed into quantitative small molecular and protein signals along with cell anatomical data into multidimensional chemical signatures which can be segmented by classification algorithms to formally classify all cells into their ultimate, natural classes (Marc et al., 1995; Marc & Jones, 2002).

In short, tissues taken post-fixation were dehydrated in graded methanols followed by acetone, then embedded in resin matrices as single specimens or stacks (Marc, 1999) followed by sectioning at 250nm into serial arrays onto 12-well HTC® slides (Erie Scientific). Sections of tissue were then probed with IgGs followed by probed CMP datasets being visualized with silver intensification (Marc et al., 1990; Kalloniatis & Fletcher, 1993; Kalloniatis et al., 1996) and counterstained with DAPI to show anatomy and simplify slice to slice registration. Images were captured in tiles at a calculated pixel scaling of 182 nm/pixel (40X planapochromatic oil immersion lens) with a QImaging QICAM IEEE-1394 12-bit cooled monochrome camera (www.qimaging.com) capturing in 8-bits. Illumination was controlled with a regulated power supply. No filtering or contrast adjustment was used prior to image analysis. Images were co-registered with irTweak, developed at the University of Utah Scientific Computing Institute (http://www.sci.utah.edu/software.html) creating n-dimensional image databases for subsequent quantitative analysis. Quantitative analysis of grey-scale images with univariate histogramming and N-space cluster analysis was implemented as described in Marc et al. (1995) and Marc & Jones (2002). Our basic tool for univariate and multivariate histogram analysis is CellKit, an application developed in the Marc laboratory under the IDL image visualization language (www.rsinc.com) which operates in the freely distributed IDL VM environment. CellKit v2.2 is available from http://prometheus.med.utah.edu/~marclab/protocols_cellkit.html. Color images were assembled with Adobe Photoshop (San Jose, California). Complete protocols are available at http://prometheus.med.utah.edu/~marclab/protocols.html.

Statistical analysis of CMP data sets

In the analysis of sample sets, all CMP procedures were performed in a masked design. All samples were coded, and the assembly of sectioning arrays, performance of thin sectioning, immunoprobing and image capture were all executed by personnel with no knowledge of the sample identity or provenance. CMP is a high-precision quantitative method (Marc et al., 1995; Marc & Cameron, 2002; Marc & Jones, 2002; Anderson et al., 2009) allowing direct comparisons of small molecule concentrations and distributions within and across samples. The signal-to-noise ratio of CMP methods is such that the probability of the classification error (P_e) < 0.001 (Marc & Liu, 2000) for a single probe and the precision is high enough to discriminate 0.04 mM differences in a single probe across cell classes (Marc & Jones, 2002). The analysis of cell signatures (i.e. combinations of small molecules) to discriminate cell classes and characterize their states was performed by K-means clustering on silver density signals from registered serial sections (as described above) and the statistical separability expressed as P_e determined by transformed divergence. A detailed introduction to the use of clustering and divergence in immunocytochemistry is provided in reference (Marc & Cameron, 2002).

Quantification of neurotransmitter staining

Signature histograms from both transplant and host retinas were acquired for GABAergic and glycinergic amacrine cells using the methods established by (Marc & Jones, 2002). The histograms were converted to cumulative frequency distributions and compared by the Komolgorov-Smirnov (K-S) non-parametric statistic.

Results

Superior colliculus (SC) responses (Figure 1; Table 1)

Figure 1. Example of visual responses in the superior colliculus.

A) Response traces to a light stimulus of 0.08 log cd/m², starting at 100 msec before and ending 300 msec after start of a 40 msec full-field light stimulus. Recordings from a normal rat (without transplant), of RD rat #5 with transplant, an RD rat with cortex transplant and an RD sham surgery rat are shown. Arrows point to the onset of visual responses. There are no responses in the RD rat with cortex transplant or with sham surgery. Transplanted RD rat #5 shows a response delayed by ~ 30 msec compared to a normal rat. B) Response maps of the right SC (oriented with rostral side up, caudal side down). Responses are indicated as black dots, non-responsive areas as circles. Normal rats have responses all over the SC. In rat #5, a small area corresponding to the placement of the transplant in the retina shows visual responses whereas normal rats have responses all over the SC and control RD rats (sham surgery, cortex transplant, non-surgery) have no responses. The response threshold of rat #5 was −2.66 log cd/m² (see Table 2).

Visual responses to a full-field light stimulus could be recorded in transplanted rats in a restricted area of the SC corresponding to the placement of the transplant in the retina (see Figure 1B). Age-matched RD rats with sham surgery or cortical transplants showed no responses. An example is shown on Figure 1. Visual response thresholds varied between −2.66 and −0.9 log cd/m² (see Table 1) with no clear correlation to transplant organization except that the rat with the least sensitive threshold contained a disorganized transplant.

Overview of statistical CMP results

In all cases described, the P_e (probability of classification error) of GABAergic amacrine cells, glycinergic amacrine cells, glutamatergic bipolar cells, and glia was <0.001. The discrimination of photoreceptors in transplant zones was absolute using the rhodopsin 1D4 monoclonal antibody, and P_e then was not computable; it was effectively 0. The key metrics in this study were the relative spatial distributions of cells and processes (bipolar cells, amacrine cells, glia) in defined transplant and host zones. Transplant zones were defined by three independent metrics: (1) regions with partial or complete outer nuclear laminations or rosettes and Rho 1D4 immunoreactivity, (2) the presence of dual inner nuclear layers, and (3) regions harboring positive BCIP signals generated exclusively by the transplant. The correlation of these three signals was perfect across all 17 samples. Thus the probability of incorrectly mapping the transplant zone in the n+1^thsample is < 1/(18)³ or 0.00017. All of the potential points of transplant GABAergic neurons fusing with the host retina occurred exclusively in these zones.

Transplant-specific NBT staining signals for hPAP in host inner plexiform layer (Figures 2- 5)

Figure 5. Electron microscopy of BCIP donor stain.

A) Transplant outer plexiform and outer nuclear layers. Irregular BCIP precipitate in outer plexiform layer, no stain of nuclei. 10 sec. uranyl acetate stain. B) Enlargement of A), showing 2 triads in outer plexiform layer. Arrow heads point to ribbon synapses. BCIP seen as irregular precipitate. C) Low magnification of transplant inner plexiform layer with BCIP stain. No counterstain. D) High magnification of transplant inner plexiform layer, different area. Arrow head points to ribbon synapse. 10 sec. uranyl acetate stain. E) Transplant-host interface. Note that BCIP stain can be seen past a lighter appearing glial membrane (black arrow). No counterstain. F) Transplant-host interface, different area in adjacent section. 10 sec. uranyl acetate stain. G) Transplant-host interface, enlargement. Note stained process past transplant-host limit. 10 sec. uranyl acetate stain. One BCIP-labeled process can be seen past the limit between transplant and host. H),I) BCIP-labeled donor derived processes in the host inner plexiform layer. Asterisks indicate unlabeled (presumable host) process in contact with BCIP-labeled (graft) process. No counterstain. A)-G),I) rat #9; H) rat #7. Scale Bars = 1 μm (A, C, E, F); 0.2 μm (B, D, G, H, I)

In sections derived from hPAP stained vibratome sections (example in Figure 3A), all the transplanted regions showed characteristic NBT diformazan densities ranging from 0.06 to 0.1 at 550 nm. This corresponds to an NBT diformazan concentration of ≈700 mM (using standard diformazan extinction coefficients and a section thickness of 200 nm), which is characteristic of diformazan reactions run to saturation. None of the peripheral host retinas outside the transplant from any of the samples showed any measurable density at any wavelength. Thus the spectral absorbance data (Figure 2) uniquely represent the presence of enzymatically reduced NBT specific for the transplant. There is no intrinsic production of the diformazan in the S344ter host retina. Further, diformazan is insoluble in aqueous media, ethanol, acetone and non-diffusible. Therefore, the reaction product in the inner plexiform layer of S344ter host retina can conservatively only be interpreted as cellular elements from the transplanted retina entering the host. Further proof is the fact that diformazan has a molecular weight of 749 Da and should readily diffuse into nuclei in the transplant region if it were diffusible. No diformazan was present in any nucleus, despite the immensely high concentration of product.

Figure 3. hPAP-stained transplant (Rat #7).

Transplant with large area of lamination. A) BCIP-stained 80 μm thick vibratome slice. B) Unprocessed density image of a 200 nm section captured at a resolution of 364 nm/pixel. This figure is an RGB composite representing the spectral absorbance at 450 (blue channel), 500 (green channel) and 550 nm (red channel). This distinguishes NBT diformazan (blue) from melanin (brown). The host retina is transparent and has no detectable signal. C) Contrast normalized nm density image: same image as A) linearly max-min contrast stretched from 8 down to 4 bits of data to reveal the shape of the infiltration region. D) shows C) with overlay of the autofluorescence image. E) Magenta (fluorescence) / green (NBT signal) overlay: green magenta map generated by inverting D) and assigning it to the R+B channels and maintaining the autofluorescence as green. Note the area of transplant processes’ penetration into the host retina (arrows). Boxes in D) and E) indicate areas of enlargement in Figure 4. In B-D), the right side of the images is tilted (deeper cut than the left side) with reduced staining (staining is strongest on the surface of the vibratome slice). Bars = 200 μm.

Figure 3 shows image processing of the BCIP/NBT diformazan signal. Proof of the unique distribution of the diformazan is provided in Figure 3B, which is an RGB (red-green-blue) composite representing the spectral absorbance at 450 (blue channel), 500 (green channel) and 550 nm (red channel). This distinguishes NBT diformazan (blue) from melanin (brown). The host retina has no detectable signal. This is an unprocessed density image captured at a resolution of 364 nm/pixel. Figure 3C is the same image linearly max-min contrast stretched from 8 down to 4 bits of data to reveal the shape of the infiltration region. Figure 3D shows Fig 3B in combination with the autofluorescence image. Figure 3E is a green magenta map generated by inverting Fig. 3D and assigning it to the R+B channels and maintaining the autofluorescence as green.

BCIP staining for hPAP identified the transplant (Figures 2-5) and indicated a very dense penetration of transplant processes into the inner plexiform layer of the host retina (arrows in Figure 3 B-D), especially at the edges of laminated transplant areas, in the transition zone to more disorganized areas (Figures 3; 4 A, B). At the EM level, BCIP appeared as an irregular precipitate (Figure 5) that densely filled the transplant areas (Figure 5 A-G). BCIP label was not found in nuclei. However, BCIP stain was only clear without counterstaining (Figure 5C, E, H, I) or with very light counterstain (Figure 5 A,B, D, F, G). The staining mostly obscured synaptic structures in the transplant although ribbon synapses could occasionally be observed (Figure 5 B, D). BCIP-labeled processes could be identified in the host inner plexiform layer (Figure 5 H, I).

Figure 4. Enlargements, showing penetration of transplant processes into the host IPL.

A) Enlargement of Figure 3 D. B) Enlargement of Figure 3 E. Boxes show areas of further enlargement in A1 and B1. Punctate stain in the host IPL is visible at higher enlargement. Bar = 50 μm.

Photoreceptors (rods and cones) (Figures 6, 7A, 8)

Figure 6. Transplant photoreceptors (rods and cones).

All images are oriented with the ganglion cell layer down. Rat #3. A) Taurine: Strong staining of transplant photoreceptors. Photoreceptors outer segments are indicated by asterisks (*). The inner retina of transplant and host both display normal taurine levels. Yellow dashes indicate approximate border between transplant and host. B) Low power micrograph. Rhodopsin staining of rod outer segments (green) can only be seen in the transplant region. DAPI stained nuclei appear magenta. White dashes indicate approximate border between transplant and host. C) Enlargement of white box in B); combination of rhodopsin (green), red-green opsin (orange), and DAPI (magenta). Arrows point to some cone outer segments among the dominant rod outer segments (only found in transplant area). D) Electron micrograph of unstained ultrathin section of BCIP stained block (Rat # 9), showing interaction between host RPE and transplant photoreceptor outer segments. BCIP can be seen as precipitate mainly in the inner segments. E) Similar area in ultrathin section of rat #7, slightly counterstained with uranylacetate. F) Enlargement of adjacent area to D) (Rat #9, stained section). Transplant outer segments in contact with microvilli of host RPE. BCIP precipitate is difficult to see because of counterstain. Bars = 50 μm (A,B); = 20 μm (C); = 2 μm (D); = 1 μm (E) = 0.2 μm (F).

Figure 7. Quantification of neurotransmitter data.

A) Normalized taurine pixel value probability frequency histograms for entire host retinal thickness (h, blue), transplant neural retina (tn, black) and transplant photoreceptor layer regions (tp, green), spanning over 2 mm of retina. Host and transplant neural retina contain indistinguishable taurine profiles due primarily to Müller and bipolar cells. The transplant photoreceptor layer contains extremely high taurine levels similar to the outer nuclear and outer segment layers of normal retina. B) Normalized GABA pixel value probability frequency (F) and cumulative frequency (C) histograms for host (h, blue) and transplant (t, black) regions, spanning over 4.6 mm of retina and encompassing the entire host and transplant inner plexiform layer. The host contained 224 and the transplant 111 GABAergic amacrine cells. Normalized for inner plexiform layer length yields 49 and 52 amacrine cells/mm of inner plexiform layer length in the host and transplant respectively. C) Normalized glutamate pixel value probability frequency histograms for host (h, blue) and transplant regions with (t1, black) and without (t2, green) photoreceptors. Glutamate signals are reliably higher in the transplant than the host and much higher in transplanted retina without photoreceptors.

Figure 8. Overview of transplant -- Glycine (red)-DAPI (green)-glutamate (blue):: red-green-blue (RGB) mapping.

Rat #3. Glycine: subpopulation of amacrine cells; glutamate: rods & cones (≈ 1 mM), BCs, GCs (5-10 mM), ACs (0.04-1 mM), MCs (0.04-0.3 mM). Reduced version of a 300 Mb image that spans one vibratome slice over approximately 4.5 mm. It is digitally broken up into segments. This section does not contain the optic nerve. The vertical arrows indicate the transplant limits and the oblique arrows indicate the limit of intact outer segments (also indicated by asterisks *). Ovals indicate patches of rod nuclei (no outer segments). 1 and 2 denote the inner plexiform layers of the transplant and host respectively. The double line indicates the region where Figures 9 A and B were taken. Bar = 0.2 mm

Transplant photoreceptors demonstrated strong taurine immunoreactivity (Figure 6A, quantification in Figure 7A). Rod and LWS (red-green opsin) positive cone outer segments in contact with host RPE could only be found in the transplant region in laminated areas (Figures 6 B,C). Transplant photoreceptor outer segments interacted with host RPE as seen in the EM (Figures 6 D, E, F). In disorganized transplant areas, there were clusters of survivor photoreceptors lacking outer segments (Figure 8). No photoreceptors with inner and outer segments could be found in the host retina.

Quantification of neurotransmitter data (Figure 7)

In no case were any significant differences in neurotransmitter level detectable between host and transplant (inner plexiform and inner nuclear layer) regions. For example, quantification of taurine levels (Figure 7A) showed that host and transplant inner retinal layers contain indistinguishable taurine profiles due primarily to Müller and bipolar cells whereas the transplant photoreceptor layer contains extremely high taurine levels similar to the outer nuclear and outer segment layers of normal retina.

Figure 7B represents the frequency and cumulative distributions for GABA signals from host and transplant GABAergic amacrine cells from the same section. The histograms are statistically identical and the K-S statistic yields P=0.46 (one-sided) which demonstrates that the distributions cannot be differentiated. One dozen comparisons were made and no transmitter showed a significant difference in level. Neuronal density was also assessed as number of neurochemical cell types per unit length of inner plexiform layer and no differences were found for GABAergic or glycinergic amacrine cells. For example, the dataset in Fig. 7B gives GABAergic amacrine cell density of 49/mm in the host and 52/mm in the transplant. Glutamate signals were reliably higher in the transplant than the host and still higher in transplant areas without photoreceptors (Figure 7C). The numbers of bipolar cells were similarly decreased in both host and transplant compared to normal retina. The remnant inner nuclear layer was significantly more variable in thickness than normal rat (30-35 μm), ranging from 10-30 μm in both host and transplant. Given that the amacrine cell frequencies seem unaltered, most of the variation must have come from bipolar cell loss. However, there were still large clusters of bipolar cells as evidenced by the high glutamate content of the host and transplant inner plexiform layers.

For reasons unknown, regions at the transplant edges that lost all photoreceptors demonstrated a much higher remnant neuronal glutamate content. In any case, the transplant regions always showed robust metabolic status to the host when photoreceptors were present.

Inner retinal neurons (Figures 8-11)

Figure 11. Glutathione, Glycine, Glutamate.

Photoreceptors outer segments are indicated by asterisks (*). A,B) Rat #3. A) Glutathione: The transplant and host both display normal glutathione levels in the inner plexiform layers, but are remodeled. Notably, the transplant has high Glutathione-+ survivor horizontal cells at the OPL/ONL border (arrows). Few stained cells in host in this area. B) Adjacent area: Glycine immunoreactivity in laminated area with outer segments (*). This area appears to contain more glycinergic cells in the transplant than in the host. C,D) Rat #1. Glutamate: magenta, DAPI: green. C) Laminated area with outer segments (*). Normal glutamate levels in transplant and host, but remodeling. Dashes indicate approximate transplant-host border. D) Host retina outside the zone of transplant. Bars = 50 μm

The transplants maintained their correct lamination in the center of the transplant with altered topologies at the edges (example in Figure 8). Both transplant and host displayed normal GABA, glutamate, glycine and glutathione levels in surviving bipolar, horizontal and amacrine cell classes (Figure 7-11), but extensive remodeling was found, especially at the transplant edges that had lost photoreceptors. The organized areas of the transplants with inner and outer segments showed the least remodeling, and showed plasticity in many areas of neuropil bridging between transplant and host (Figure 10 D, F).

Figure 10. GABA-ergic amacrine cells.

GABA: green, DAPI (nuclei): Magenta. Dashes indicate approximate transplant-host border. A,B) Rat #1. Photoreceptors outer segments are indicated by asterisks (*). A) Overview. GABA stain reveals remodeling, especially in areas with photoreceptor loss at the edge (not shown) and outside the transplant. Area of enlargement is indicated by white box. B) Enlargement of transplant. C) Host retina outside transplant area, rat #2. No outer nuclear layer. D) Rat #1, different slice than in A,B. Clear process mixing and fusion of migration columns. E) Rat #2. No obvious mixing. F) Different slice of rat #2 with transplant GABA neurons remodeling (migrating) into the host zone. Bars = 500 μm (A); 50 μm (B, D, E, F); 20 μm (C)

Figure 8 shows a combination of Glycine-DAPI-glutamate RGB mapping through the extent of a vibratome slice. The transplants appeared to maintain a relatively uniform thickness, but the transplant photoreceptor layer and inner nuclear layer thickness was inversely related to the inner plexiform layer thickness. The transplant showed loss of inner retinal neurons, specifically bipolar cells. The inner nuclear layer of the transplant was overall thinner than that of the host (Figure 9A). Areas of bridging between transplant-host neuropil was seen at the edges of laminated areas (Figure 9B), and also in areas with two inner plexiform layers (Figure 10 E,G). GABA staining (Figure 10) showed extensive remodeling, especially at the transplant edges and in the host retina outside the transplant (Figure 10C). Transplant and host displayed normal GABA levels (Figure 7B). Although there were many areas with no or sparse mixing of transplant host-neuropil (examples in Figures 10B, F), apparent mixing of transplant-host neuropil could be observed in the glycine, GABA and glutamate channels (Figure 9, 10E, G). Transplants contained many glutathione-immunoreactive horizontal cells (Figure 11A) and there were usually more glycinergic amacrine cells in the laminated transplant area than in the degenerated host retina (Figure 11B). Glutamate labeling (Figures 8, 9, 11C, D) revealed areas of transplant photoreceptors and patches of rod nuclei within the transplant regions. Glutamate levels appeared normal in both the transplant and host inner retinas (slightly elevated in the transplant) (Figure 7C), but as in glycine labeling, showed remodeling events by glutamatergic neuronal populations.

Figure 9. Enlargements -- Glycine (red)-DAPI (green)-glutamate (blue).

Two areas of Figure 8, rat #3. Photoreceptors outer segments are indicated by asterisks (*). A). Laminated area. 1 and 2 denote the inner plexiform layers of the transplant and host respectively. The inner nuclear layer of the transplant is overall thinner than that of the host. It appears still thinner in areas where the inner plexiform layer of the transplant is thicker, and vice versa. B) Higher magnification view of the underlined region in the Figure 8, at the edge on the laminated area. The left RGB image is mirrored by the right glycine channel alone (intensity mapped like fluorescence). The transplant-host neuropil is mixed all the way from the microneuroma at the top, though the transplant, into the host. Bars = 50 μm

Glial cells (Figure 12)

Figure 12. Glial markers – areas without inner limiting membrane between transplant and host.

Asterisks (*) indicate presence of outer segments. A) Rat #3. CRALBP (marker for Müller glia and RPE) is present but less expressed in the transplant area with photoreceptors. CRALBP is much stronger in the remodeled, more disorganized transplant areas (not shown on this picture). B)-D) Rat #5. Transplant area showing some disorganization, imaged with combinations of GABA staining (green) with various other labels. B) DAPI: magenta. Arrows indicate transplant-host interface. B₁) Enlargement. C) CRALBP: magenta. C₁) Enlargement. D) CRALBP: red; Glutamine synthetase (GS): blue. Glial seal on left side (arrowhead in (C) and (D)) is interrupted through most of the area, with neural processes crossing the transplant-host interface. Bars = 50 μm

CRALBP was present at lower concentrations in the transplant zone with photoreceptors (Figure 12A), but much stronger in the remodeled transplant zone (not shown). A glial seal revealed by CRALBP and glutamine synthetase labeling was observed between the host and transplant retina (Figure 12 C,D). However, there were many regions where the seal was clearly absent revealing areas of bridging between host and transplant retinas (detail shown in Figure 12 B₁, C₁). Additionally, transplant and host both displayed normal glutamine levels, but glial glutamine levels were elevated in the remodeled zones at the transplant edges and in the host retina outside the transplant (data not shown).

Discussion

After transplantation to retinal degeneration models using a gentle transplantation method, fetal retinal sheet transplants have been shown to develop retinal layers and then retain their lamination, including photoreceptor outer segments (Seiler & Aramant, 1998), to form synaptic connections with the host (Seiler et al., 2008c; Seiler et al., 2010) and to restore visual responses in the superior colliculus (Sagdullaev et al., 2003; Thomas et al., 2004; Yang et al., 2010). CMP has uniquely revealed the rearrangement of retinal topology and demonstrated circuit revision (retinal remodeling) in photoreceptor degenerations (Marc & Jones, 2003; Marc et al., 2003; Jones & Marc, 2005; Marc et al., 2007; Marc, 2010). This study sought to implement CMP (Marc et al., 1995; Marc & Jones, 2002) to visualize cell classes and to examine potential circuitries involved in fetal retinal sheet transplants.

1. Visual responses in the superior colliculus (SC)

Electrophysiological recording from the SC was used to provide additional information about the function of the transplants. In normal rats, responses were found all over the SC at different light intensities down to −5 log cd/m². No visual responses could be recorded in age-matched degenerate RD retina controls at the tested light levels of −0.08 to −3.5 log cd/m² whereas responses with longer response latencies were found in transplanted rats in a small area corresponding to the placement of the transplant in the host retina. Visual thresholds varied between transplanted rats. The mechanism of restoration of visual function has been explained as a rescue effect of the transplant on host cones by some research groups. E.g., MacLaren et al. (MacLaren & Pearson, 2007) stated that visual responses in transplanted rats could be explained by a mere rescue effect of host cones. However, previous research has shown (Sagdullaev et al., 2003; Thomas et al., 2004; Seiler et al., 2008b; Seiler et al., 2008c; Seiler et al., 2010) that it is not a rescue effect although a beneficial trophic effect of the developing fetal transplant on the host is likely involved. After retinal sheet transplantation in this S334ter line 3 rat model, there was no difference in the distribution of red-green opsin immunoreactive cones in the host retina over the transplant area compared to outside the transplant whereas visual responses in the SC were limited to areas that topographically corresponded to the transplant position in the host retina (Seiler et al., 2008b). In addition, we have demonstrated that the degree of trans-synaptic labeling of the transplants from the SC is correlated with the visual threshold in SC recordings (Seiler et al., 2008c), and that transplants extend neuronal processes into the inner plexiform layer of the host retina and form synapses with host cells (Seiler et al., 2010).

2. Photoreceptors

In all examined samples, cell survival and retinal remodeling appeared in a mosaicked fashion with the transplant containing areas of healthy rods and cones whereas the photoreceptors in the host retina were degenerated at the time points observed. Cone opsin immunoreactivity in the transplants was similar to normal rats while rod or cone opsin in the host retina was not observed. Other studies have shown surviving host cones using red-green opsin (Seiler et al., 2008b; Yang et al., 2010) or S-antigen/arrestin immunohistochemistry (Sagdullaev et al., 2003). Although S334ter-3 rats never develop rod outer segments (Liu et al., 1999), cone outer segments - labeled with peanut agglutinin - are found over the whole retina at P10, diminish at P20-30, and are reduced to isolated spots at P60-90 (Li et al., 2010; Martinez-Navarrete et al., 2011). At P180, S-opsin immunoreactive cones show amacrine cell-like morphology and entirely lack outer segments (Hombrebueno et al., 2010).

The current study implies that degenerating host cones are not capable of any light response. Correspondingly, visual responses in the SC were restricted to an area corresponding to the placement of the transplant in the retina, confirming previous studies (Sagdullaev et al., 2003; Seiler et al., 2008b; Yang et al., 2010).

3. Bipolar cells

In every sample, fewer inner retinal neurons were found in the transplants compared to the adjacent host. There appeared to be an inverse relationship between the thickness of the transplant photoreceptor layer and the transplant inner plexiform layer.

In the host and to a lesser extent in the transplant, an extensive loss of inner retinal neurons, specifically bipolar cells (BCs) was observed. More than 90% of the cells in the transplant inner nuclear layer consisted of amacrine cells (ACs) and Müller cells (MCs). In the normal retina, the inner nuclear layer contains more bipolar than amacrine and Müller cells together (BCs > MCs = ACs) (Jeon et al., 1998).

Bipolar cells were likely lost in host and graft due to cell death (Marc et al., 2003; Ray et al., 2010), or they did not properly develop in the transplant (as in a Chx−/− mouse) (Livne-Bar et al., 2006).

In S334ter-3 rats, a significant reduction (50-75%) of rod BCs has been observed at ages P60-90 (Ray et al., 2010). Similar to other retinal degeneration models (Jones et al., 2003; Marc & Jones, 2003; Peng et al., 2003; Cuenca et al., 2004; Strettoi et al., 2004), rod BCs in S334ter-3 rats respond to photoreceptor loss with abnormal sprouting and ectopic synapses (Ray et al., 2010). However, numerous rod BCs have been shown by PKC alpha and mGluR6 immunohistochemistry in fetal retinal sheet transplants, (e.g., Seiler et al., 2008b). Since the total number of BCs was reduced with likely no loss in rod BCs, this indicates therefore a specific loss of cone BCs in the transplant inner nuclear layer. It corresponds to unpublished data indicating that transplants contain few recoverin-immunoreactive cone bipolar cells. This will need to be further explored as the current study was not designed to solve this issue.

At the time of transplantation (E19), the donor tissue consisted mostly of retinal progenitor cells, except for ganglion cells, cones, some amacrine and horizontal cells (Uesugi et al., 1992; Alexiades & Cepko, 1997). Rods and rod BCs are derived from the same progenitor (Turner & Cepko, 1987), and compete for their cell fate choice (Brzezinski et al., 2010) which can occur after the final mitosis (Belliveau et al., 2000). The bHLH genes Mash1 and Math3, and the homeobox gene Chx10/Vsx2 are essential for generation of bipolar cells (Hatakeyama et al., 2001). The degenerating host retina may present a different environment for the retinal progenitor cells in the transplant than a normally developing fetal retina.

4. Glial seal

Photoreceptor degeneration is accompanied by extensive hypertrophy of Müller cells and development of glial seals (Jones et al., 2003).

It has been postulated that retinal transplants cannot integrate with the host retina because of a glial seal between transplant and host which needs to be disrupted for any connectivity to occur (Suzuki et al., 2007; West et al., 2008). However, the current study confirms that fetal retinal sheet transplants can “integrate” with a degenerating host retina, in areas lacking a glial barrier between transplant and host (e.g., Seiler & Aramant, 1998; Seiler et al., 2010). The transplants were performed at a time when all host rod photoreceptors were lost, but still at an early stage of degeneration. The absence of the glial seal could be due to the influence of the developing transplant to inhibit the development of a glial seal formation or to disruption of the outer limiting membrane due to the surgery. The BCIP staining for the donor cell label indicated that transplants extended processes into the host retina, confirming previous studies (Seiler et al., 2010).

In addition, novel connectivity between a degenerating retinal host and an embryonic transplant may depend on the model of retinal degeneration as transplants to rd mice demonstrated less host/transplant integration (Arai et al., 2004). In subretinal transplants to normal retinas or to slow retinal degenerative models, the outer limiting membrane surrounding photoreceptors may function as a barrier for transplant host integration (Zhang et al., 2003).

5. Remodeling

Early in retinal degenerations, bipolar cells retract dendrites from the photoreceptor terminals and subsequently form polyaxonal processes that sprout into the ONL and potentially also into the IPL. These bipolar cell processes also contribute to the formation of new neuropil along with amacrine and horizontal cells (Jones & Marc, 2005). These new processes could also be responsible for the integration of neuropil between host and transplant; and the inner plexiform layer demonstrated in the distorted IPL images of the transplant in this study.

Retinal remodeling of the transplant was more pronounced than that of the host retina with apparent neovascularization outside the photoreceptor layer. Retinal remodeling represents the likely recapitulation of mechanisms that resemble development and plasticities seen elsewhere in the CNS (Jones & Marc, 2005; Draguhn et al., 2008). It had previously been hypothesized that retinal transplants would also remodel (Jones & Marc, 2005; Jones et al., 2005; Marc et al., 2005), but that there were no mechanisms in place to establish communication between transplanted retina and the degenerating host retina. Indeed, the concern was also that there were no mechanisms in place to prevent transplant remodeling. However, this study demonstrates that plasticity-like phenomena could serve as a means for retinal transplant integration into the host retina. Neurons within the host and transplant retinas appear to represent healthy cell populations and it may be possible to influence their emergent rewiring and cellular migration to guide future transplant integrations.

The current study could not establish whether there were any differences between the BDNF and GDNF treatment of the transplants due to the low N and transplant variability. A more detailed analysis of transplant functional effects after BDNF or GDNF treatment has been published recently (Yang et al., 2010).

6. Novel non-canonical circuitry

Given that this is a rodent model, cone BCs are rare in both the host and transplant. In spite of cone bipolar cell loss in the transplant, there were still more cone BCs in the transplant than in the host along with substantial populations of surviving horizontal cells visualized with glutathione immunoreactivity. This finding, along with previous results (data not shown) suggests that horizontal cells contribute substantially to retinal remodeling and may represent a target for study and intervention with fetal transplants. Other obvious contributors demonstrated by the glycine and GABA channels (Figs 9, 10, 11) suggest that any major communication between the host and transplant is likely to involve horizontal and both glycinergic and GABAergic amacrine cells (ACs).

The data confirm studies of trans-synaptic tracing (Seiler et al., 2005; Seiler et al., 2008c) and identification of donor-derived synapses in the host IPL (Seiler et al., 2010) demonstrating that synaptic connectivity between transplant and host contributes to restoration of visual responses in the SC. Further study is required to determine the retinal circuit topology between host and transplant retinas. However, this knowledge will be dependent upon a more complete understanding of normal retinal circuitry which is currently grossly limited (Anderson et al., 2009; Anderson et al., 2011).

Conclusions

At the time of examination (about 4 – 10 months after transplantation), retinal progenitor sheet transplants contained regions of healthy rods and cones superimposed upon rod/cone degeneration in the host retina. Extensive loss of neurons, including bipolar cells, was observed in the transplant inner nuclear layer although remaining cell populations appeared healthy. Frequently, host and transplant neuropil had become mixed where glial barriers were missing. Horizontal cells, and both glycinergic and GABAergic amacrine cells are likely involved in bridging connectivities between transplant and host, potentially creating alternative signaling pathways. While these studies give substantial support to vision therapy strategies involving sheets of retinal progenitor cells, more work is needed to define transplant/host integration in retinal degenerative disease.

List of abbreviations used in figures and text

AC

amacrine cell

BC

bipolar cell

BCIP

5-Bromo-4-chloro-3-indolyl phosphate (alkaline phosphatase substrate)

CMP

computational molecular phenotyping

CRALBP

cellular retinaldehyde binding protein

DAPI

4′6-diamidino-2-phenylindole dihydrochloride

GABA

gamma-aminobutyric acid

GC

ganglion cell

GS

glutamine synthetase

hPAP

human placental alkaline phosphatase (donor label)

INL

inner nuclear layer

IPL

inner plexiform layer

LWS

long wavelength sensitive

MC

Müller cell

NBT

Nitro blue tetrazolium chloride (alkaline phosphatase substrate)

ONL

outer nuclear layer

OPL

outer plexiform layer

OS

outer segments

RD

retinal degenerate or retinal degeneration

RGB

red-green-blue

RPE

retinal pigment epithelium

SC

superior colliculus