Retinofugal Projections from Melanopsin-Expressing Retinal Ganglion Cells Revealed by Intraocular Injections of Cre-Dependent Virus

To understand visual functions mediated by intrinsically photosensitive melanopsin-expressing retinal ganglion cells (mRGCs), it is important to elucidate axonal projections from these cells into the brain. Initial studies reported that melanopsin is expressed only in retinal ganglion cells within the eye. However, recent studies in Opn4-Cre mice revealed Cre-mediated marker expression in multiple brain areas. These discoveries complicate the use of melanopsin-driven genetic labeling techniques to identify retinofugal projections specifically from mRGCs. To restrict labeling to mRGCs, we developed a recombinant adeno-associated virus (AAV) carrying a Cre-dependent reporter (human placental alkaline phosphatase) that was injected into the vitreous of Opn4-Cre mouse eyes. The labeling observed in the brain of these mice was necessarily restricted specifically to retinofugal projections from mRGCs in the injected eye. We found that mRGCs innervate multiple nuclei in the basal forebrain, hypothalamus, amygdala, thalamus and midbrain. Midline structures tended to be bilaterally innervated, whereas the lateral structures received mostly contralateral innervation. As validation of our approach, we found projection patterns largely corresponded with previously published results; however, we have also identified a few novel targets. Our discovery of projections to the central amygdala suggests a possible direct neural pathway for aversive responses to light in neonates. In addition, projections to the accessory optic system suggest that mRGCs play a direct role in visual tracking, responses that were previously attributed to other classes of retinal ganglion cells. Moreover, projections to the zona incerta raise the possibility that mRGCs could regulate visceral and sensory functions. However, additional studies are needed to investigate the actual photosensitivity of mRGCs that project to the different brain areas. Also, there is a concern of "overlabeling" with very sensitive reporters that uncover low levels of expression. Light-evoked signaling from these cells must be shown to be of sufficient sensitivity to elicit physiologically relevant responses.


Introduction
Melanopsin-expressing retinal ganglion cells (mRGCs) in the eye have been recently recognized as important mediators of non-image forming visual responses, such as circadian photoentrainment and pupillary light responses, in many mammalian species [1]. Published maps of central projections from mRGCs [2][3][4][5] have provided an important groundwork for formulating hypotheses related to the physiological and behavioral responses modulated by mRGCs. These previous studies relied on indirect labeling of mRGCs with an antibody against pituitary adenylate cyclase-activating peptide [2], or on low titer of intravitreal AAV-GFP [3], or on the direct labeling of all melanopsin-expressing cells using a reporter gene (Opn4-lacZ; [4]) or the genetic Cre-lox system (Opn4 cre ::AP loxP and Opn4 cre ::GFP loxP ; [5,6]). The Cre-lox based genetic approach has been the most sensitive technique to reveal the diversity of mRGCs subtypes and their central targets. However, this approach also revealed extra retinal expression of floxed reporter genes in cells across many brain areas of Opn4 cre mice including cerebral cortex, thalamus and brainstem [5] that are not thought to express melanopsin. Melanopsin has also been recently found in the iris of mice using immunohistochemistry [7]. These findings complicate the use of melanopsin-driven genetic labeling techniques to identify retinofugal projections specifically from mRGCs In this present study, we employed an alternative genetic method to specifically label mRGCs in the eye and to further increase the labeling of mRGCs with weak melanopsin expression. Delivery of a Cre-dependent reporter using a recombinant adeno-associated virus (AAV) has become a popular tool to label genetically defined cells. This method offers two advantages over the systemic Cre-lox genetic reporter approach. First, it does not reveal the historic pattern of expression thereby eliminating the report of transient expression in cells during early development. Second, it delivers multiple copies of the Cre-dependent reporter per infected cell thereby increasing the chance of recombination in cells with weak Cre expression. Therefore, we decided to further investigate the central targets of mRGCs in the brain by intravitreal injection of AAV carrying floxed human placental alkaline phosphatase (PLAP) reporter into Opn4 cre mice [5]. Here we present the results of these tracing studies.

Generating AAV-flex-plap
This virus was generated using standard sub-cloning with a modified pAAV-MCS backbone. The cDNA encoding human placental alkaline phosphatase (PLAP, NM_001632) was flanked by loxP ( Fig 1A, open triangles) and lox2722 ( Fig 1A, closed triangles) sites to yield the flexplap transgene. This transgene was inserted in reverse orientation into a modified pAAV-MCS plasmid 3' to a CMV promoter and 5' to a woodchuck hepatitis virus post-transcriptional regulatory element (WPRE) and bovine growth hormone polyadenylation (pA) sequence to generate pAAV-flex-plap. High titer virus of serotype 2/1 (4x10 12 IU/mL) was generated from this plasmid at the University of North Carolina, Chapel Hill Vector Core facility.

Animals
Mice were housed in an AALAC-accredited pathogen-free animal facility with ad libitum access to food and water and with a 12-hour light-dark cycle with lights on at 7AM and off at 7PM. The University of California, San Francisco Institutional Animal Care and Use Committee (IACUC) specifically approved this study. The protocols, animal care procedures and the experimental methods meet all of the guidelines on the care and use of laboratory animals by the U.S. Public Health Service.

Intravitreal injections
The age of the mice ranged from P38 to P96 at the time of injection. Mice were anesthetized with Isoflurane and topical administration of proparacaine (0.5%; Bausch & Lomb). The pupils were dilated by topical administration of phenylephrine (2.5%; Bausch & Lomb) and atropine sulfate (1%; Bausch & Lomb) eye drops. A 32-gauge Hamilton syringe was used to inject 2 microliters of AAV-flex-plap into the superior part of the vitreous of right eyes. A total of 13 injections were made (11 into Opn4 cre mice and 2 into C57BL/6J wild-type mice). No PLAP signal was detected in the retinas of wild-type mice. Visually detectible PLAP labeling was examined and quantified in the brains of 5 animals.
Tissue processing for PLAP histochemistry PLAP histochemistry was performed as previously described [9,10]. Two to eight weeks after the intravitreal injection the mice were euthanized by CO 2 and transcardially perfused with 10 ml HEPES-buffered saline (HBS; 8.2 g/l NaCl, 6 g/l HEPES, 0.1 g/l Na 2 HPO 4 , pH to 7.4 with NaOH) followed by 20 ml cold 4% paraformaldehyde (PFA) in HBS. All subsequent solutions were prepared with HBS unless otherwise noted. Brains and eyes were dissected and post-fixed in 4% PFA for 3-4 hours at 4°C. The brains were embedded in 3% agar and cut on a vibratome (Model 3000, Vibratome Company) into 100 μm sections. The retina was dissected from the fixed eyes and flat-mounted. Following rinsing in HBS, the endogenous alkaline phosphatase was heat-inactivated by incubation at 72°C for 1 hour. Tissue was then washed twice in buffer 1 (100 mM Tris pH 7.5, 150 mM NaCl) and then twice again in buffer 2 (100 mM Tris pH 9.5, 100 mM NaCl, 50 mM MgCl 2 ). PLAP reporter was visualized by incubating tissue in buffer two with BCIP (5-bromo-4-chloro-3-indolyl phosphate, 0.2 mg/ml) and NBT (nitro blue tetrazolium, 1 mg/ml) at room temperature for 1 to 12 hours. PLAP reaction was monitored and stopped before background staining became excessive by rinsing 3 times in 1mM EDTA followed by post-fixation in 4% PFA for 1 hour. To remove background staining, the tissue was cleared by immersing in ethanol series (30%-70%-95%-100%-95%-70%-30%) for 1-3 minutes in each series and the subsequent wash in the HBS. Brain sections were counterstained (see below). Brain sections and whole mount retinas were mounted on microscope slides, briefly rinsed in distilled water and cover slipped using Aqua/Poly mount (Polysciences, Cat. # 18606).

Counterstaining
To visualize brain nuclei, all brain sections were counterstained with ToPro3 (Life Technologies, Cat. # T3605), a fluorescent nuclear stain, at 1:5,000 dilutions. To better visualize thalamic nuclei, alternate sections were processed for Cytochrome Oxidase staining [11] by immersing slices in Cytochrome Oxidase staining solution (30 mg Cytochrome C, 20 mg Catalase, 50 mg DAB in 100 ml PBS) for 2-4 hours at room temperature.

Labeling melanopsin retinal ganglion cells
We previously reported light-driven, melanopsin-dependent activation of neurons in the central amygdala and posterior thalamus in neonatal mice using immunolocalization of immediate early gene expression [12]. To further understand the contribution of melanopsin-expressing retinal ganglion cells (mRGCs) in the eye to non-image forming responses, we decided to look more carefully at projections of mRGCs to these and other brain areas. When we crossed Opn4 cre mice [5] with mice carrying floxed tdTomato reporter [8], we found tdTomato in neurons of the inner retina, presumed to be mRGCs. We also found tdTomato in many areas of the brain that are not known to be direct targets of retinal ganglion cells, such as the somatosensory cortex and cerebellum (S1 Fig). Similar extra retinal expression of Opn4 cre driven reporter was also observed by Ecker et al. [5].
To limit the expression of Opn4 cre driven reporter to retinofugal projections from mRGCs and to maximize the labeling of mRGCs with weak melanopsin expression, we created an adeno-associated virus (AAV) vector with floxed human placental alkaline phosphatase gene (AAV-flex-plap, Fig 1A; see methods). The intravitreal injection of AAV-flex-plap into Opn4 cre mice leads to robust labeling of many mRGCs around the site of injection ( Fig 1B). The signal was always specific to the injected eye and was never observed in the contralateral, non-injected eye ( Fig 1C). This result confirms that the virus did not spread outside the injected eye and that the labeling is specific to the virally encoded PLAP and not a result of endogenous alkaline phosphatase activity. We found that control injections of AAV-flex-plap into the vitreous of wild-type mice did not result in any labeling (n = 2, data not shown) thereby confirming that the PLAP signal is specific to Opn4 cre expressing retinal cells.

Brain areas innervated by retinal mRGCs
Two to eight weeks after the intravitreal injection of AAV-flex-plap, the brains were sectioned coronally and processed to visualize PLAP. A representative coronal series is presented in Figs 2 and 3. The following sections describe our findings (summarized in Tables 1 and 2). For each case analyzed the entire series of sections was scored for the presence or absence of PLAP-positive axons in each brain area. Adjacent sections processed for cytochrome oxidase staining and ToPro3 counterstains were used to identify architectonic boundaries in the sections. The Mouse Brain in Stereotaxic Coordinates [13] and Allen Mouse Brain Atlas [14] were used to identify regions and nuclei in the stained sections. For AOS anatomy we also referred to Giolli et al. [15].
Accessory optic system and the superior colliculus. Another novel finding was the observation of dense contralateral innervation by mRGCs of the accessory optic nuclei including dorsal, lateral, and medial terminal nuclei (DT, LT, and MT 5/5; Fig 4). The superior colliculus (SC, 5/5) received dense, mostly contralateral, innervation (Fig 4). Central targets of mRGCs in the brain. Representative coronal sections (rostral to caudal) from the brains of Opn4 cre mice with intravitreal injection of AAV-flex-plap into the right eye. Left panels in each row are low-power images of brain sections processed to visualize alkaline phosphate. The next panel in each row is an image of an adjacent brain section stained for Cytochrome Oxidase. Some insets are composite montages of several images (D, E, F). See Table 1 for nomenclature. Scale bar: 200 μm (inset, 100 μm). Ipsilateral and contralateral sides of the brains are labeled. Note that HDB and MPA areas appear on the "contralateral", right side of the section. Since these sections are rostral to the chiasm, one might think fibers in HDB and MPA should be on the ipsilateral side. However, we have traced these fibers back to the chiasm and find that they have indeed crossed over from the ipsilateral side and projected in the rostral direction.
doi:10.1371/journal.pone.0149501.g002 Fig 3. Central targets of mRGCs in the brain. Representative coronal sections (rostral to caudal) from the brains of Opn4 cre mice with intravitreal injection of AAV-flex-plap into the right eye. Left panels in each row are low-power images of brain sections processed to visualize alkaline phosphate. The next panel in each row is an image of an adjacent brain section stained for Cytochrome Oxidase. See Table 1

Discussion
In the present study, we used viral delivery of a Cre-dependent reporter to label melanopsinexpressing retinal ganglion cells (mRGCs) in the mouse eye and to determine what central targets in the brain they innervate. We found that mRGCs innervate multiple nuclei in the basal forebrain, hypothalamus, amygdala, thalamus and midbrain. Midline structures tended to be bilaterally innervated whereas the lateral structures received mostly contralateral innervation. These results are in accord with the previously published results [2][3][4][5]. The following sections highlight the novel targets and discrepancies with the previous studies. A caveat of the intravitreal viral delivery of the reporter is that this technique does not label all mRGCs in the eye. It labels only a subset of mRGCs near the site of injection (Fig 1B). Therefore, this technique only samples the projections by mRGCs from one region of the retina; there might be additional brain regions innervated by mRGCs that were not labeled by our injections. Another caveat is that PLAP labeling in various brain areas does not necessarily represent synapses in these regions; PLAP labeled fibers could represent fibers of passage. A critical question is whether we were viewing axon terminals targeted to different brain regions or might the PLAP-positive fibers be axons of passage? Based on our impression that the axons ended in a particular region we assumed this was an area targeted by mRGCs. However to substantiate this assertion we would have to do high-resolution microscopy and antibody labeling. Axon terminals should be enriched for synaptic proteins such as VGLUT2 or synapsin, or show structural specializations usually associated with synaptic boutons.

Central amygdala
A novel finding of this study is sparse but consistent innervation of the central amygdala ( Fig  2E and 2F). These findings suggest that the light-induced neural activation in CeA of the neonatal mice [12] may be due to direct activation of neurons in CeA by mRGCs. The extent of direct vs. indirect effect of mRGCs on neural activation in CeA remains unknown. We speculate that if this retina-amygdala pathway is conserved in humans, it may be the neural mechanism of aversive responses to light in neonatal and preterm infants [16,17].

Zona incerta
We found consistent projections of mRGCs to the zona incerta (Fig 3B-3E). It is one of the limbic nodes and is implicated in the regulation of water and food intake, reproductive behaviors and cardiovascular activity [18]. Additionally, the zona incerta is involved in processing pain by controlling the transmission of signals from the spinothalamic tract to the posterior thalamus [19]. In primates, zona incerta is also implicated in controlling saccades via GABAergic projections to the superior colliculus [20]. Altogether, our findings suggest that mRGCs may play role in regulating these non-image forming responses. Posterior thalamus An electrophysiological study by Noseda et al. [21] of light-responsive neurons in LD, LP, VP and Po in rats suggested a direct input from mRGCs to these thalamic regions. We were able to detect projections from mRGCs to LD and LP but not to VP or Po (Fig 3A-3E) in mice. The observed lack of innervation in VP and Po does not necessarily prevent neurons in these brain areas from receiving light signals originating from mRGCs. The dendritic fields of VP and Po neurons are large enough (350 μm; [22]) to receive direct visual input from nearby visual thalamic areas that are heavily innervated by mRGCs. Additionally, the absence of direct innervation of Po and VP that we report here could reflect differences between mice and rats.

Accessory optic system (AOS)
An unexpected finding of our study is the heavy innervation of the accessory optic system by mRGCs. We saw dense projections to all AOS nuclei including OT, DT and MT (Figs 3E and 4) suggesting that mRGCs contribute to tracking responses. Consistent with this suggestion, an M4 subtype of mRGCs was recently found to contribute to the contrast sensitivity of the optokinetic response [23]. However, in contrast to our findings, Hoxd10 retinal ganglion cells, which include many types of AOS projecting directionally selective cells in the retina [24], do not express melanopsin and are not intrinsically photosensitive as determined by immunolabeling with anti-melanopsin antibody and whole-cell electrophysiology (personal communication with Maureen Stabio and David Berson). Additionally, the previous study by Ecker et al. [5] using the same Opn4 cre mouse strain and a genetic reporter did not detect any signal in the AOS. One possible resolution of the apparent disparity between our results and that of Ecker et al. is that AAV-mediated delivery of Cre-dependent reporter may result in multiple copies of viral genomes per infected cell, thereby increasing the chance of detecting cells with low level of melanopsin expression. It is possible that this increased sensitivity revealed novel AOS projecting subtypes of mRGCs that were not previously described. There are a number of additional experiments that need be carried out to further investigate the contribution of mRGCs to visual tracking: (a) analysis of labeled mRGCs to see if they include directionally selective retinal ganglion cells; and (b) retrograde labeling of retinal ganglion cells projecting to MT and OT nuclei followed by analysis of melanopsin expression and of their intrinsic photosensitivity.

Retinal photoreceptors
We found examples of PLAP positive cones in pAAV-flex-plap transfected retinas (S2 Fig). These results corroborate previous studies reporting melanopsin-driven reporter expression in photoreceptors (Ecker et al. 2010). A critical question from a developmental standpoint is when is Opn4 active? In this present study we find PLAP expression in retinas injected after postnatal day 30. Whether melanopsin photopigment contributes to detection of light in the Opn4-expressing photoreceptors will require further study.
Moreover, our ability to observe PLAP-positive photoreceptors is informative with respect to finding an AAV serotype (2/1) that is capable of infecting photoreceptors when injected into the vitreous. This has importance for devising gene therapy interventions in outer retinal degenerative diseases. Viral transfection of rods and cones following an injection different AAV serotypes into the vitreous is uncommon. Of 7 different serotypes tested by Hellstrom et al. [25] only AAV2/3 and AAV2/5 seemed efficacious for transducing rods and cones. Our findings show that AAV2/1 can also transduce photoreceptors, suggesting AAV2/1 has properties similar to AAV2/3 or AAV2/5. Further study is required to quantitatively compare transduction percentages of these different serotypes.
Two issues using AAV based markers deserve consideration. 1. An important concern that arises from the use of highly sensitive reporters is the relationship between the labeling and the physiologically relevant level of melanopsin expression. As labeling techniques become progressively more sensitive, with an ability to mark ever-lower levels of expression, a question arises as at what point the labeling becomes excessive and is not physiologically relevant. It is possible that our technique labeled RGCs with low levels of melanopsin expression that may not be sufficient to confer intrinsic photosensitivity or a significant biological function. Further studies are needed to show that the labeled mRGCs are indeed intrinsically photosensitive.
2. An important question that arises is whether the virus we used transfected selective classes of mRGC. The intersectional genetic strategy used by Ecker et al. [5] showed that Opn4-cre was virtually ubiquitously expressed, however, we cannot assess whether our PLAP-flex-plap virus infected all subtypes of mRGC. Differences in the density of staining in target regions identified by intravitreal injections of a rAAV-GFP (Gooley, et al, [3]) and ours is consistent with the notion that different viruses may not infect all classes of ipRGC. Hellstrom et al. [25] illustrated evidence for selective transduction of subsets of single classes of retinal cell by different AAV serotypes. In conclusion we cannot say whether selective classes of ipRGC were transfected by our AAV-flex-plap injections.
In summary, our findings highlight sparse but diffuse innervation of multiple limbic areas in the brain suggesting that light activation of mRGCs may contribute to regulation of multiple homeostatic responses in the animal including sleep, feeding, reproductive behaviors, cardiovascular function, alertness, mood, pain and memory. Additionally, projections to the accessory optic system suggest that mRGCs play role in saccades and visual tracking, responses that were previously attributed to other classes of retinal ganglion cells.