Neuropsin-expressing cells in the retina affect melanopsin expression and response to light in mice

In our efforts to characterize retinal cells expressing neuropsin, we employed the OPN5-Cre strain⁶. Through intravitreal injection of AAV carrying a Cre-dependent fluorescent reporter, we achieved the specific labeling of cells expressing neuropsin (OPN5, Fig. 1A) and immunolabelled melanopsin expressing cells (OPN4, Fig. 1B).

OPN5-expressing cells were found distributed throughout the entire retina, with no discernible specific pattern of distribution (as depicted in Fig. 1A). Notably, many of the labeled neurons exhibit dendritic and neuronal features, strongly indicating their identity as retinal ganglion cells (RGCs). This characterization was further confirmed through examination of retinal cryosections (as shown in Fig. 1C). Importantly, all neurons labeled by the fluorescent reporter were RGCs on all sections imaged. We only observed one occurrence of a labelled cell outside of the GCL. This cell was observed in the INL, possibly a displaced RGC. In addition, we did not observe any cell stained for both neuropsin and melanopsin. We measured soma sized of melanopsin- and neuropsin-expressing cells and established that their soma diameter distribution greatly varies (Fig. 1D). In average, melanopsin-expressing cells have larger soma than neuropsin-expressing cells (melanopsin: 26.16 ± 0.19 μm; neuropsin: 10.65 ± 0.06 μm). We counted a density of 146.9 ± 16.3 OPN5-positive cells per mm² with AAV labeling in OPN5^Cre/+ retinas (Fig. 1E).

We immunolabeled OPN4 in OPN5^Cre/+ and OPN5^Cre/Cre retinas to compare the density of OPN5-positive neurons with mRGCs and assess the impact of the absence of OPN5 on mRGC population. Indeed, in OPN5-Cre strain, the Cre-recombinase was inserted in place of Opn5 exon 1, erasing the expression of OPN5 protein⁶. OPN5^Cre/Cre are thus deficient in OPN5. We observed a significatively lower number of OPN4-positive neurons in these OPN5^Cre/Cre retinas (OPN5^Cre/+: 109.09 ± 5.93 mRGCs/mm², OPN5^Cre/Cre: 82.72 ± 4.22 mRGCs/mm²; p = 0.036; Fig. 1E). To note, it has been shown that immunostaining of OPN4 only labels mRGCs subtypes with the highest expression of OPN4, M1-M3 ³. It is, however, unclear if the absence of neuropsin directly impacts mRGCs intrinsic light response, melanopsin expression, or if it is a consequence of retinal vasculature overgrow previously identified in absence of neuropsin⁶. Overgrowth of blood vessels can cause increased leakage and leads to defect of visual functions^13,14.

To identify the bilateral projections to the brain of OPN5-expressing cells, we injected AAV expressing Cre-dependent EGFP (left eye) and AAV expressing Cre-dependent tdTomato (right eye) in OPN5^Cre/+ mice eyes. This approach extensively labels dendrites, soma as well as axonal processes in Cre-expressing neurons. Axons from OPN5-positive neurons exhibit diverse projections to multiple brain regions associated with visual functions. A few axonal branches were observed within both the core and shell regions of the hypothalamic SCN that serves as the master circadian oscillator (Fig. 2A). However, the OPN5 positive axons in the SCN are notably sparser than the mRGC axons¹⁵ that serve as primary conduit of light input to the SCN. Beyond the hypothalamus, OPN5 axons project to the ventral (vLGN) and dorsal (dLGN) lateral geniculate nucleus, with a clear segregation of axons of ipsi- and contralateral origin (Fig. 2B). The dLGN is a relay to the visual cortex and involved in the scene perception¹⁶ while vLGN has recently been describe to modulate response to threats¹⁷. Contralateral neuropsin cell axonal terminals are concentrated in the dorsal part of the dLGN, which has been described as innervated mainly by direction-sensitive RGCs¹⁸, and the external vLGN, the retinorecipient part of the vLGN¹⁹. The ipsilateral terminals in both vLGN and dLGN are localized in the central patches. No overlapping between ipsi- and contralateral terminals are observed in these structures. The intergenicular lamina (IGL), localized between vLGN and dLGN, appears to be spared by OPN5-expressing axons terminals. In the thalamic region, the olivary pretectum nucleus (OPN), implicated in pupil constriction, on the other hand, shows substantial overlap of bilateral axons from OPN5-expressing cells (Fig. 2B). Although the OPN harbors synaptic terminals from RGCs, including mRGCs¹⁵, the mRGC axonal processes are known to be more numerous than other RGCs in the OPN. Ipsilateral terminals are concentrated in the shell of the OPN while contralateral terminals are mainly in the core of the OPN, with some overlap in the lateral part of the shell. Finally, the axons of neuropsin-expressing cells project to the nucleus of the optic tract (NOT), the superior colliculus (SC) and the accessory nuclei (MT) (Fig. 2B, C). These regions are involved in gaze-stabilization and in image forming. These axons predominantly display a contralateral pattern of innervation within these structures. This pattern of projection further underscores the potential involvement of neuropsin-expressing cells in visual processing (Fig. 2C). In summary, the results suggest a distinctive role for OPN5-expressing cells in these neural circuits and potentially highlights their unique contributions to visual and circadian processes.

Characterizing RGC subtypes based solely on morphological parameters poses a considerable challenge and demands a level of resolution adequate to distinguish the dendritic arborizations of each RGC subtype. This task becomes particularly difficult with OPN5-expressing RGCs, as observed in light microscopy because their dendrites largely overlap (Fig. 1A). The limitations of light microscopy highlight the necessity for electron microscopy, which not only provides the essential resolution but also grants access to the intracellular content of examined cells, including mitochondrial populations and synaptic sites.

Employing advanced techniques like serial blockface scanning electron microscopy (SBEM), coupled with specific EM staining for OPN5 cells, allowed us to reconstruct and identify these RGC subtypes. To produce EM-grade OPN5-specific staining of RGCs, we intra-vitreally injected 2 AAVs to express Cre-dependant EM reporter APEX2²⁰ and Cre-dependant fluorescent marker tdTomato to a male OPN5^Cre/+ mouse. After 4 weeks, the retina was fixed and processed to reveal APEX2 labeling. APEX2 and tdTomato signals were compared to verify the efficiency of the labeling. We then chose a region of the peripheral retina containing multiple stained RGCs somas to image in EM. The final SBEM 3D image volume consisted of 948 sections with each covering 113.2 μm x 113.9 μm area with inter-section interval of 50 nm. The volume contained a total of 90 neuronal soma on a surface of 12,893 μm² (~ 7500 neurons/mm²). The image volume extended from the GCL to the inner limit of the INL, for a total volume of about 600,000 µm³ (Fig. 3A). A total of 4 neuropsin-expressing neurons were identified by their darkened cell membrane (Fig. 3B) marked by APEX2 labeling. Their neurites were also labeled and appeared almost completely dark (Fig. 3C), thus allowing us to easily follow the neurites through the sections. The dendrites regularly present swellings containing mitochondria and form synapses with nearby axons. The other organelles of the cells were not labeled by APEX2.

Our initial step involved the automated segmentation of APEX2-labeled dendrites. We achieved this segmentation by leveraging the visual distinction between these dendrites and other cells. Our method involved a straightforward pixel value detection technique, the details of which can be found in the Methods section. In brief, the threshold detection applied to the whole volume created 1,000,038 contours matching our pixel intensity and size thresholds. The IMOD functions imodmesh and imodsortsurf sorted these contours based on their interpolated 3D surfaces into 132,312 objects. Using a custom Python script, we removed all objects smaller than 15 sections to eliminate small false-positive elements such as lysosomes or mitochondria. The number of objects was reduced to 3000 that way. Finally, we manually checked the remaining objects to verify they are all dendrites and merged objects corresponding to the same cell (Supplementary Video 1). To evaluate the quality of the prediction, we manually check each object created and establish the false-positive rate (FPR) in the GCL and the IPL (Supplementary Fig. 1). The GCL had a relatively high FPR (71%), due to large structures dense to electrons which escape our filtering steps, such as lysosomes, nucleoli or Muller cells cytoplasm. In the IPL, the FPR was notably lower (41%); the structures falsely segmented were processes of Muller cells. This pipeline of semi-automatic segmentation allows a fast morphological characterization of cells with dark labelled neurites and can be easily included in connectomics studies workflow.

This segmentation process successfully identified the presence of four APEX2-labeled cells, with three of them exhibiting dendrites within the imaged volume. Additionally, a dense network of dendrites was observed throughout the entire inner plexiform layer (IPL) (Fig. 4A). These dendrites covered a large portion of the retina surface (Fig. 4B). While proofreading the segmentation, we however discovered that some dendrites were lightly labelled and, although clearly distinguishable from unlabeled neurites, were not automatically segmented (Fig. 4C).

To ensure a comprehensive analysis of dendritic arborization for these cells and avoid the risk of missing any dendrites, we carried out a manual refinement of the autosegmentation process using VAST-lite software²¹. This involved systematically tracking and fully segmenting all APEX2 labeled dendrites until they either reached the volume's edge or came to an end. As anticipated for retinal ganglion cells (RGCs), the dendrites exhibited distinct organization into specific sublayers of the inner plexiform layer (IPL). To our surprise, the dendrites of all three cells displayed unique branching patterns within the IPL. Specifically, OPN5 neuron 2 had dendrites predominantly terminating in the outer region of the ON sublayer of the IPL, while OPN5 neuron 1 exhibited termination in the OFF sublayer of the IPL. OPN5 neuron 3 had dendrites that extended into both the ON and OFF sublayers of the IPL. This distinctive segregation of dendritic arborization patterns suggests a specialization of these RGCs in processing specific types of visual information in the retina. (Fig. 5A). The distribution of dendrites within each sublayer of the retina was quantified to assess the relative density of dendrites (Fig. 5B). Some dendrites of all three neurons went out of the volume, making any measurement of their dendritic radius impossible. However, we were able to estimate the relative distribution of these dendrites in relation to their respective soma using a classical analysis for neurons morphology, the Sholl analysis^22,23. Because of the dendrites continuing outside of the volume, the Sholl analysis would be only partial but the dendrites present in the volume were sufficient to establish the difference between the neurons. Indeed, we obtained very different profiles for the three neuropsin-expressing neurons. Neuron 2 has mostly branches close to the soma while neuron 1 dendrites spread farther away (Fig. 5C). We also quantified the content of each soma but a meaningful comparison is difficult as the volume starts from the middle of the soma. We thus estimated the portion of the visible soma occupied by the nucleus and mitochondria. All data are presented in Table 1.

Next, we calculated what fraction of the retina surface may be covered by OPN5 expressing cells. The dendrites of the three neuropsin-expressing cells collectively covered a substantial portion of the retinal surface, accounting for approximately 30.6% of the total retinal surface. However, it's worth noting that within the imaged volume, there were numerous other dendrites expressing OPN5, which did not originate from these three cells. The initial result from the automated segmentation indicated that around 42% of this volume surface was covered by dendrites from OPN5-expressing neurons.

Recognizing that this percentage might be underestimated due to unsegmented dendrites, we adopted a strategy to obtain a more accurate representation. We partitioned the volume into 25 sections based on a 5 by 5 grid and randomly selected five of these sections for the full segmentation of all labeled dendrites. This approach allows us to gain a better understanding of the overall coverage and distribution of OPN5-expressing dendrites in the retina. (Fig. 6A). Dendrites were distributed in all layers of the IPL, with a preference for the OFF sublayer of the IPL and very few dendrites in the inner part of the ON sublayer, possibly proximal segments of dendrites from the somas of OPN5-expressing neurons (Fig. 6B). This distribution is consistent with the three cell subtypes we identified earlier, even though only a small portion of the dendrites traced in the boxes overlap with the dendrites from those 3 labeled OPN5-expressing neurons. Using this method, we confirmed that the OPN5-expressing dendrites covered an average of 42% of the retina surface in the current image volume, ranging from 32% to 58% (Fig. 6B).

We then assessed whether all layers of the retina contained OPN5-dendrites receiving synapses. We marked all synapses in one of the previously segmented boxes and compared the distribution of synapses to the distribution of dendrites across the sublayers of INL (Fig. 6C). We counted a total of 108 synapses distributed in proportion to the density of dendrites in the corresponding box (Fig. 6D). The inner ON sublayer had the lowest number of dendrites and synapses, while the OFF sub-layer had the highest density of dendrites and synapses. Thus, OPN5-cell dendrites receive synaptic inputs from all sublayers of the INL.

OPN5 is presumed to be a UV-sensitive opsin²⁶. To verify if UV light stimulation of mouse retina can lead to an electrical response of neuropsin-expressing cells, we recorded light responses of the ganglion cell layer of the retinas of OPN4^Cre/Cre mice using multi electrode array (MEA). As these retinas lack melanopsin, the observed photoresponses should originate from the outer retina photoreceptors or from OPN5 neurons. We first recorded the retina piece perfused with aCSF only and applied four UV light stimulation of increasing irradiance (365 nm, from 1.17 10¹¹ to 1.02 10¹⁴ photons/cm²/s) (Fig. 7A). Around 40% of the active cells showed a dose-depending response (Fig. 7B). To eliminate the response due to SW-cones, we then incubated the retinas in a cocktail of synaptic blockers for one hour. The same light protocol was then applied again to the retina. With the synaptic blockers, no response could be detected (Fig. 7C). A previous study showed that neuropsin cells might depend on the exogenous supply of retinal to respond to light²⁴, we, therefore, incubated the retinas with the cocktail of synaptic blockers and 11-cis-retinal for 1 h before applying the same light protocol again. No light response was detected either (Fig. 7D). Finally, to verify that the absence of response was not due to a survival problem of the retina, retinas were washed with fresh aCSF, without synaptic blockers. After 30 min of washing, partial restoration of the light response was detected upon application of the light stimulations (Fig. 7E). Six retina pieces from 3 different OPN4^Cre/Cre mice were recorded and exposed to the same light protocol with the same absence of response in the presence of synaptic blockers. This suggests that OPN5-expressing cells do not have an electrical response to light in these conditions.

We subsequently evaluated if the absence of OPN5 would also impact the light response of mRGCs. We recorded OPN5^Cre/Cre mouse retinas, which are deficient in neuropsin, on a MEA and exposed them to increasing intensity light stimulations (15 s, from 4.82 10¹¹ to 9.01 10¹⁵ photons/cm²/s, Fig. 8A). We stimulated with 470 nm monochromatic light, close to the peak of the sensitivity of OPN4²⁵ and used five stimulations to show a better resolution of the response. We obtained the expected intensity-dependent response from mRGCs from both WT and OPN5^Cre/Cre retinas (Fig. 8B). We however observed a reduced average discharge rate in OPN5^Cre/Cre at the highest irradiance (WT: 3.91 ± 0.60 spikes/s, n = 107; OPN5^Cre/Cre: 2.35 ± 0.25 spikes/s, n = 19; p < 0.05; Fig. 8C) without any significant change to the total duration of the response (WT: 22.90 ± 5.86 s; OPN5^Cre/Cre: 25.63 ± 3.54 s; p = 0.67; Fig. 8D).

All animal care and procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of the Salk Institute for Biological Studies and in accordance to ARRIVE guidelines. All the methods were carried out in accordance with relevant guidelines and regulations. A mouse strain expressing Cre-recombinase under the control of Opn5 promoter (Opn5^Cre) was obtained from Dr. Richard Lang from Cincinnati Children's Hospital Medical Center⁶. The OPN4-Cre and OPN5-Cre strain were maintained using heterozygotes animals as breeders. All experiments comparing genotypes were done with littermates. For histology and SBEM, male and female Opn5^Cre/+ of 6–12 weeks of age were used. For electrophysiological recording, homozygotes OPN4^Cre/Cre and OPN5^Cre/Cre mice of 8 and 12 weeks of age were used. All mice were housed under a standard 12:12 light/dark cycle (~ 100 lx at cage level) in a temperature-controlled room (~ 22 °C). Food and water were available ad libitum.

To label neuropsin cells with the fluorescent proteins EGFP and tdTomato, AAV2-EF1a-DIO-EGFPf and AAV2-EF1a-DIO-tdTomato were injected into, respectively, the left and right eye of Opn5^Cre/+ mice. To label neuropsin-expressing cells for electron microscopy, we produced a vector with a farnesyl sequence cloned into the 3' end of the APEX2 construct²⁰ and inserted it in an inverted orientation between the lox sites in an AAV2-DIO vector³⁴. This viral vector, AAV2-EF1a-DIO-APEX2-farnesyl, was co-injected with AAV2-EF1a-DIO-tdTomato into both eyes of Opn5^Cre/+ mice to fluorescently label APEX2 expressing cells with tdTomato, so that these cells could be identified under a fluorescent microscope. All viral vectors were produced by the Salk Gene Transfer, Targeting and Therapeutics Viral Vector Core Facility at titers of 2.68 × 10¹¹ GC/mL (APEX2-f), 9.43 × 10¹¹ GC/mL (tdTomato), and 2.12 × 10¹² GC/mL (EGFP-f).

Anesthesia was induced with isoflurane (4%) and maintained (2%) until the end of the procedure. The mouse was placed on one side under a dissection microscope so one eye was completely visible. Gentle pressure was applied around the eye so the edge of the sclera was visible. A small incision was made with a 27-gauge insulin needle 0.5 mm posterior to the Ora Serrata. The viral vector was loaded into a Hamilton microliter syringe with a 34-gauge beveled needle mounted on a micromanipulator. The micromanipulator was used to insert the loaded needle through the incision. The vector was slowly injected and allowed to diffuse through the vitreous humor for 2 min. The whole procedure was then repeated on the other eye with the appropriate vector. Upon delivery of the viral vector to both eyes, the animal was removed from isoflurane anesthesia and lubricant eye ointment (AKORN) was applied to both eyes. The animal was placed in a clean cage to recover and was returned to its home cage after 1–2 min when righting reflex was restored.

We automatically detected and traced dendrites marked with APEX2 that appear dark in EM images using the imodauto function of the software IMOD[³⁶]. The function used image thresholding to segment objects. Here, we detected all pixel values under a fixed threshold corresponding to the average value of the labeled dendrites. If the pixels detected formed a large enough area, they are saved as a contour. Once all images of the volume are processed, we meshed all contours to create 3D surface with contours overlapping over multiple sections (imodmesh). The surfaces were then separated by different objects (imodsortsurf). We then used a custom Python script to delete all objects that were smaller than 15 sections to eliminate small structures such as lysosomes that also appear dark in EM images. Lastly, all remaining objects in every section were manually checked to confirm that they only contained labeled dendrites.

Results were expressed as means ± standard error of the mean (SEM). Comparisons of the two groups were done using a non-parametric rank-sum test (Mann-Whitney U Test). Comparisons of multiple groups were done by using ANOVA on ranks (Kruskal-Wallis H test) followed by a post hoc test (Pairwise Mann-Whitney U-tests) with Benjamini-Hochberg p-value correction. A statistically significant difference was assumed with a corrected p-value inferior to 0.05. All statistical analyses were performed using R (R project). All custom scripts generated in this study are available on GitHub (https://github.com/Carboneinerte↗).