A Cre Mouse Line for Probing Irradiance- and Direction-Encoding Retinal Networks

Small subsets of amacrine and ganglion cells are marked in the Rbp4-Cre mouse

We revealed Cre expression in retinas of Rbp4-Cre mice by injections of Cre-dependent mCherry virus (AAV) or crosses with a tdTomato Cre-reporter strain (Ai14). In both cases, red fluorescence marked a minority of RGCs and ACs. BCs were also sparsely labeled, along with a subset of photoreceptors, but we have not studied these further.

In one fully mapped retina from an Rbp4-Cre;Ai14 mouse (Fig. 1A,B), there were 1893 tdTomato-positive cells in the GCL, corresponding to a mean density of 115 cells/mm² (retinal area = 16.4 mm²). The density of labeled cells showed modest topographic dependence, being lowest in the dorsonasal retina (Fig. 1C). Labeled cells in the GCL comprised a mixture of RGCs and ACs. The RGCs were identifiable from their clearly labeled axons, traceable into the optic fiber layer, and from their immunoreactivity for the pan-ganglion-cell marker RBPMS (Rodriguez et al., 2014; Fig. 1D–F). However, most fluorescent cells of the GCL lacked both of these features and were thus displaced ACs. Approximately 90% of the fluorescent cells of the GCL (∼1700 cells in total) were displaced ACs, while the remainder (∼200 cells in total) were RGCs, as determined by RBPMS staining. The relative abundance of these cell classes varied topographically, with RGCs making up only 2.5% of the labeled cells in the ventrotemporal GCL, but 14.6% of them nasally (Fig. 1D–G).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Density of Rbp4 ACs and RGCs varies across the retina. A, Photographic montages of a representative retinal wholemount from a Rbp4-Cre;Ai14 mouse showing tdTomato labeling in Cre-expressing cell bodies in the GCL (black dots). Fluorescence image has been inverted and binarized for clarity. D, dorsal; N, nasal; V, ventral; T, temporal. B, Magnified view of retinal region marked by black square in A. tdTomato fluorescence is pseudocolored magenta; focal plane is in the ganglion-cell layer. C, Density of tdTomato-positive cells in the GCL varies topographically and is lowest in the dorsal-nasal retina. Values for each quadrant are mean densities (cells/mm²) derived from counts in four fields, each 256 × 256 μm. D–F, tdTomato-labeled cells of the GCL (D) include RGCs (white circles) identifiable by their RBPMS-immunolabeling (E) and displaced ACs (yellow arrows), which are RBPMS immunonegative. G, Topographic variation in the percentage of tdTomato-positive cells that are RGCs.

Labeled RGCs were readily divisible into two distinct types. The first, more common type (here termed the R-cell) was bistratified with a relatively small soma and compact dendritic field (Fig. 2A,B). The second, rarer type (here termed the RDS cell) was also bistratified, but resembled classic DS cells. These had distinctly larger somas and, typically, asymmetric dendritic arbors (Fig. 2A,C). An additional AC type was labeled in this line. These had small somata in the conventional position for ACs (inner margin of INL) and a bistratified dendritic arbor. Figure 2D compares the patterns of stratification for each of these four types, and a fuller description of each follows.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Morphology of Rbp4-Cre-positive cells of the inner retina. A–C, Rbp4-Cre-positive cell types of the GCL. All images are maximum intensity projections showing the arbors as seen en face. To highlight the structure of the injected cell, images have been masked to hide extraneous fluorescence and pseudocolored to distinguish between processes in the ON sublayer of the IPL (green) from those in the OFF sublayer (purple). All cells were filled by intracellular dye injection after targeted recording of a tdTomato-positive RGC or AC in the Rbp4-Cre-tdTomato mouse. A, Wide-field Rbp4-Cre-labeled displaced AC (RAC), shown to scale with the two RGC types, the RDS-cell and R-cell. B, R-cell. C, RDS-cell. Arrowheads in B, C indicate axons. All scale bars, 50 μm. D, Schematic diagrams illustrating stratification of Rbp4-Cre-labeled ACs (left) and RGCs (right) relative to ipRGCs and starburst ACs (ON and OFF ChAT bands). Polyaxonal displaced ACs (RACs, cyan) restrict their processes to the inner ipRGC plexus (lower green band). Bistratified ACs (purple), which have conventionally placed somas in the INL, stratify within both the outer ipRGC plexus (upper green band; the accessory ON sublayer, locus of ectopic synapses from ON cone BCs) and in the inner ipRGC plexus. R-cells (red), the most abundant Rbp4-Cre-labeled RGC type, deploy an outer arbor between the OFF ChAT band and the outer ipRGC plexus; their inner arbor stratifies near the inner boundary of the ON ChAT band. RDS-cells (blue) deploy their dendrites strictly within the ON and OFF ChAT bands.

Rbp4-Cre labels a unique population of displaced polyaxonal ACs of the inner ON sublayer

For convenience, we refer to Rpb4-Cre-positive displaced ACs (i.e., the labeled ACs in the GCL) as RACs. Their morphology is summarized in Figures 2A,D, 3 and Table 1. Cell bodies were relatively large for ACs (11.2 ± 1.4 μm in diameter; average ± SD; n = 46), comparable to those of neighboring starburst ACs and the smallest RGCs. They were multipolar and extended their processes horizontally within the inner ON sublayer of the IPL (Fig. 3A). There, they were quickly lost in a dense feltwork of labeled processes dominated by very fine varicose fibers following mostly straight trajectories (Fig. 3A, magenta, B, red). This plexus occupied the inner third of the ON sublamina (S4/S5), entirely proximal to (below) the ON cholinergic band. This plexus was matched almost exactly in depth with the inner plexus of melanopsin immunoreactive processes (Fig. 2D, left), consisting mainly of the dendrites of M2 ipRGCs (Fig. 3C–F; Berson et al., 2010). Intracellular dye injections of Cre-labeled RACs revealed their morphology in better detail (Figs. 2A and 3A, green, G). Their processes spanned roughly 1mm, encompassing about a fifth of the retina each (Fig. 3G; diameters 970 ± 130 μm; n = 4). The most widely spreading processes appeared axonal in form, having a uniform very fine caliber with periodic swellings, but most primary processes seemed dendritic, with a gradually tapering caliber as well as some spines and appendages. We therefore conclude that RACs are a form of wide-field polyaxonal AC (Fig. 2A).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Morphology of ACs labeled in the Rbp4-Cre mouse line. A–G, Morphology of RACs, the Rbp4-Cre-positive wide-field polyaxonal displaced ACs. RAC processes form a dense plexus in the inner ON sublayer intermingled with dendrites of M2 ipRGCs. A, Single intracellularly dye-filled RAC (green) in relation to the dense plexus of fine, horizontally running tdTomato-positive processes in the Rbp4-Cre;Ai14 mouse (magenta). Maximum intensity projection of several optical sections through the GCL and inner ON sublayer of the IPL. B, Relationship of the same plexus of wide-field ACs processes (red, tdTomato reporter for Rbp4-Cre) with the dendritic plexus of M2 ipRGCs (blue, anti-melanopsin). Tiled maximum intensity projection of ∼10 optical sections spanning the plexus and adjacent GCL but entirely proximal to ON ChAT. Scale bar, 50 μm. C, Magnified subregion from B showing numerous points of potential contact between the two cell types. Scale bar, 10 μm. D–F, Orthogonal projections of processes of Cre-positive ACs of the GCL (D), inner and outer ipRGC dendritic plexuses, marked by anti-melanopsin immunofluorescence (E), and a merged view (F) showing costratification of amacrine plexus with that of M2 melanopsin dendrites. G, Drawings of single dye-filled RACs. H–J, Cre-expressing wide-field bistratified ACs of the INL. H, Somata and outer plexus of processes of these cells as shown in maximum intensity projection of a shallow z-stack of confocal optical sections spanning the INL-IPL border. I, Single optical section at the level of this plexus in the most distal IPL showing the processes of tdTomato-labeled AC processes (red) with the dendrites of M1 ipRGCs (cyan), labeled by anti-melanopsin immunofluorescence. Somata of OFF starburst ACs (green) are also visible at this level. J, Orthogonal maximal intensity projection of the z-stack from H, showing the two plexuses of Rbp4-Cre-labeled ACs, an outer one (above) derived from the INL ACs (arrows) and the inner one (below) derived from both these INL ACs and RACs. Scattered labeling between these plexuses comprise connecting dendrites of the INL ACs. Somata in the GCL are not visible because the z-stack spanned only the IPL and proximal INL.

Rbp4-Cre labels a sparse population of conventionally placed bistratified ACs

A second type of Cre-labeled AC in this mouse line had conventionally placed somata in the INL (Figs. 2D purple, 3H–J). They were regularly spaced, suggesting possible mosaic organization (Fig. 3H), and of uniformly small size (diameter of 9.65 ± 0.55 μm, average ± SD, n = 23). Cre-dependent viral labeling revealed that their processes form a dense plexus in the outermost IPL, just proximal to their somata (Fig. 3H–J). Well-labeled cells almost invariably possessed a major dendritic process that descended through the IPL and branched, and apparently arborized, within the plexus of Cre-labeled RAC processes in the inner ON sublamina (Figs. 2D, left, 3D–F). We could not determine from such material the dimensions of the ON-sublayer dendritic fields of these bistratified ACs, nor what fraction of this inner amacrine-cell plexus was derived from these cells as opposed to RACs.

RACs have sustained irradiance-encoding light responses

A recent study in rat retina found that all ON sustained wide-field displaced ACs are electrically coupled to ipRGCs (Reifler et al., 2015). The RACs we identify among the labeled Rbp4-Cre cells are similar morphologically to these rat ACs, so we suspected that RACs might have similar functional properties. Indeed, whole-cell current-clamp recordings confirmed this inference. Cre-labeled RACs were tonically depolarized, and increased their spiking in response to prolonged whole-field illumination (Fig. 4A). Responses included a transient peak that decayed within a few seconds to the steady-state level. Firing rates during this plateau phase varied as a function of stimulus irradiance. This is evident from spiking visible in individual voltage traces (Fig. 4A), from the average PSTH (average ± SEM; Fig. 4B), and from the Irradiance-Response (IR) function of pooled cells (Fig. 4C), derived from spike frequency during the five last seconds of the stimulus. We also examined the excitatory currents by voltage-clamping at the chloride reversal potential. Sustained excitatory inward currents persisted throughout the stimulus, with amplitudes correlated with stimulus irradiance (Fig. 4D,E). Light offset evoked an outward current (Fig. 4D) and modest hyperpolarization (Fig. 4A). These cells exhibited little evidence of an antagonistic surround (data not shown).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

RACs have sustained irradiance-encoding light responses even under blockade of chemical synapses. A, Voltage responses, including spiking, of a single RAC stimulated with a series whole-field light steps of various irradiances under control conditions (Ames’ medium). Whole-cell current-clamp recordings. Irradiance levels in log photons cm⁻² s ^{− 1} are indicated above traces. Horizontal black bar below the leftmost trace indicates the timing of the 15-s stimulus (spot diameter, 1200 µm). B, Average PSTHs of spiking evoked by light steps of various irradiances (n = 3 cells; average ± SEM). C, IR curve based on data presented in B. Data points and error bars indicate the average ± SEM of the steady state response, assessed during the last 5 s of the stimulus. Magenta line denotes the fitted sigmoidal Naka-Rushton function. The coefficient of determination (R ²) for the fit is indicated. D, Excitatory inward currents evoked by a series of stimulus irradiance steps under control conditions. Whole-cell recording with voltage clamped at the chloride reversal potential. E, IR curve derived from data in D. F–I, Effect of ON-channel blockade (bath applied L-AP4) on the voltage responses of RACs to light steps. Whole-cell current-clamp recordings. F–I, Voltage responses of a representative RAC during ON-channel blockade. F, Response to the brightest light step tested (black trace; 13.5 log photons cm⁻² s ^{− 1}). Lower trace is approximately to scale with those in A. Inset, Higher gain display showing transients at onset and offset of the stimulus, plus a slow depolarizing drift of the membrane potential during the stimulus. The lower envelope of the voltage trace is shown in magenta. G, Voltage responses at several irradiances (n = 3 cells; average ± SEM). Lower voltage envelope is shown, rather than raw voltage, to minimize distortion by spikes. H, IR relationship based on data in G. I, Voltage responses to a series of four bright wide-field light steps each 30 s long. Note the gradual depolarization during the stimulus and slow poststimulus decay, which are characteristic of intrinsic melanopsin photoresponses in ipRGCs. J–M, Effects of combined OFF and ON channel blockade on RAC light responses obtained in whole-cell current-clamp configuration (bath applied L-AP4 plus an OFF cocktail that included D-AP5, DNQX, and ACET; n = 3 cells), conventions the same as for matching plots in F–I. Light still drives a slow depolarization, presumably due to persistence of melanopsin-dependent photoresponses in ipRGCs.

To probe the origin of these light-evoked inward currents, we blocked the ON pathway by bath application of L-AP4 (an agonist of the Group III metabotropic glutamate receptors essential for light responses in ON BCs). ON-channel blockade dramatically reduced the light-evoked depolarization and firing (compare Fig. 4F, lower trace and Fig. 4A), but low amplitude responses persisted (Fig. 4F, inset). For bright stimuli (>10¹² photons cm⁻² s ^{− 1}), these consisted of a rapid hyperpolarization (∼1 mV), followed by a larger, slowly developing sustained depolarization (Fig. 4F,G,I) that lasted at least 30 s, and decayed over many seconds after stimulus offset (Fig. 4I). Dimmer stimuli evoked only a very small sustained hyperpolarization (Fig. 4G). At light intensities above threshold for the sluggish depolarization, response amplitude increased with stimulus irradiance (Fig. 4G,H). This response is strikingly similar to melanopsin-driven responses of ipRGCs; it is depolarizing and irradiance encoding, with a high threshold (∼10¹² photons cm⁻² s⁻¹), sluggish onset and slow poststimulus decay (Berson et al., 2002). Because these ACs are melanopsin-immunonegative (Figs. 3B, 5G), our findings parallel those of Reifler et al. (2015) in suggesting that the sluggish depolarizing response seen under these recording conditions derives from electrical coupling to ipRGCs.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

RACs are tracer coupled to M2 ipRGCs and other neurons of the GCL. A–C, Intracellular dye filling and tracer coupling of a Rbp4-Cre displaced AC (RAC) in relation to melanopsin immunoreactivity. Montage of collapsed z-stacks (maximum intensity projections) covering a portion of the RAC’s arbor. A, Fluorescent streptavidin labeling of Neurobiotin-injected RAC (purple circle; partly obscured by artifacts) and in eight cells tracer coupled to it (white and gold circles). Magenta arrowheads indicate the extremely slender processes of the dye-filled RAC. B, Anti-melanopsin immunolabeling. A tracing of the filled RAC is overlaid in white. Circles around tracer-coupled cells are color coded for the melanopsin immunoreactivity of the marked cell: gold for immunopositive cells and white for immunenegative ones. C, Merged view of A, B. D–I, Identity of tracer coupled neurons in region marked by upper white square in A--C. Conventions as in A–C. D, Neurobiotin tracer. E, Anti-melanopsin immunofluorescence. F, Cre-dependent tdTomato labeling. G, Merge of D–F. H, Immunofluorescence for the RGC marker RBPMS. I, Merge of Neurobiotin (D) and RBPMS labeling (H). White arrowhead indicates a tracer-coupled cell that is both RBPMS immunenegative, and thus an AC, and Cre-negative and thus not a RAC. J–O, Same cell as A–I but for the retinal region marked by the lower white square in A–C. P, Tracing of the Neurobiotin-injected RAC, which nearly encompasses this retinal quadrant; retinal margin marked by blue line. Q, Drawings based on dendritic anti-melanopsin immunelabeling of three partially reconstructed M2 ipRGCs (green) that were tracer coupled to the Neurobiotin-filled RAC (black profile). Also traced are two M1 ipRGCs (red) in the same field that were not tracer coupled to the injected RAC.

RACs also appear to get modest excitatory input from the OFF channel. Blockade of the ON channel revealed a modest brisk depolarization at light offset (Fig. 4F,G,I). This was blocked (Fig. 4K,M) by supplementing the ON blocker L-AP4 with a mix of drugs designed to block the OFF channel. This “OFF cocktail” included D-AP5 (an NMDA glutamate receptor antagonist), DNQX (an AMPA/kainate glutamate receptor antagonist), and ACET (a selective GluR5-containing kainate receptor antagonist). The sluggish, irradiance-encoding, melanopsin-like response persisted under this more extensive blockade of chemical synaptic signaling (Fig. 4J,K,M).

RACs are tracer coupled to selected ipRGCs and ACs

We used tracer coupling to test the hypothesis that the melanopsin-like response seen in RACs is attributable to gap junctional coupling to ipRGCs (Reifler et al., 2015). In Rbp4-Cre;Ai14 mice, two fluorescently tagged RACs were injected with Neurobiotin, which passes through gap junctions to label coupled cells. Both injected cells were confirmed to be wide-field ACs, and one of these is illustrated in Figure 5. Tracer-coupled cells labeled in these experiments included M2 ipRGCs, identified by their melanopsin immunoreactivity and characteristic dendritic arbors (Fig. 5B,E,K,Q). Also tracer-labeled were displaced ACs, identified by their small soma size, and their immunonegativity for either melanopsin or the RGC marker RBPMS (Fig. 5H,I,N,O). These ACs were not RACs, because they were not labeled by the tdTomato Cre reporter (Fig. 5F,G,L,M). A few tracer-coupled cells were melanopsin-immunonegative and weakly RBPMS immunopositive. It is unclear whether these are ipRGCs with weak melanopsin expression (M4 or M5 cells) or some other RGC types. We conclude that RACs form an extensive electrically coupled network comprising, at least, M2 ipRGCs and a distinct type of displaced ACs.

Rbp4-Cre marks two types of bistratified RGCs

To assess the general morphology of Cre-expressing RGCs in the Rbp4-Cre line, we labeled large numbers of them through genetic crosses with the Ai14 tdTomato Cre-reporter mice or by Cre-dependent viral labeling. Melanopsin and ChAT immunolabeling provided laminar benchmarks. We could relatively easily trace the coarser dendrites of the RGCs through the thicket of amacrine-cell processes in the inner ON sublayer and thus reconstruct their arbors with reasonable confidence. Two clear types emerged: the smaller R-cell (Fig. 2B) and the larger RDS cell (Fig. 2C). Both were bistratified, but they differed in many other respects, as outlined below.

Morphology of the R-cell

These RGCs were invariably bistratified. Their somas were small (diameter, 12.9 ± 1.0 μm); not much larger than those of starburst ACs, and smaller than those of M1 ipRGCs (Fig. 6F). Their dendritic fields were of intermediate size among murine RGCs, and their ON arbor (178 ± 35 μm in diameter) was typically slightly larger and more complex than the OFF arbor (138 ± 36 μm; Figs. 2B, 6A,C; Table 1). Although the two arbors were usually largely in spatial register, they were substantially offset from one another in a minority of cells. Local clusters of labeled R-cells had the appearance of fragmentary mosaics with well-spaced somas and a uniform plexus of moderately overlapping dendritic arbors in both the ON and OFF sublayers of the IPL (Fig. 6F,G). One tier of dendrites lay within the ON IPL sublayer, stratifying near the proximal margin of the ON ChAT band. These processes lay just distal to those of M2 ipRGCs and the wide-field displaced RACs described above (Fig. 6D). The ON tier loosely cofasciculated with the ON starburst processes (Fig. 6E). The OFF arbor was rather broadly stratified, filling the space between the plexus of OFF starburst AC dendrites and the plexus of M1 ipRGC dendrites marking the INL-IPL border (Figs. 2D, right, 6D).

• Download figure
• Open in new tab
• Download powerpoint

Figure 6.

Morphology of the R-cell, a bistratified RGC labeled in the Rbp4-Cre line. A, Examples of the common tdTomato-positive RGCs, R-cell, in the Rbp4-Cre mouse. B, Variation in soma diameter across cell types in the GCL. Soma diameter differed significantly across the three cell types of the GCL [one-way ANOVA, F_(2,65) = 90.9, p < 0.001; post hoc pair-wise comparisons revealed that each cell type differed from the other two (p < 0.001)]. C, Dendritic field diameter as a function of number of branch points for RACs, R-cells and RDS-cells. D, Stratification of R-cell processes. Side-view maximum intensity projections of processes in Rbp4-cre cells (red) in relation to processes of M1 and M2 ipRGCs (purple; anti-melanopsin) and to starburst AC (SAC) processes (green; anti-ChAT). The fine Rbp4-cre-positive processes at the level of the inner melanopsin plexus (M2 ipRGC dendrites) are derived from wide-field displaced RACs; the coarse processes are derived mainly from R-cells, and stratify near and within the ON ChAT band and between OFF ChAT and the outer melanopsin plexus (M1 ipRGC dendrites). E, Dendrites of the ON arbor of two virally labeled R-cells (green; flex AAV2-mCherry) tend to cofasciculate with the processes of ON starburst ACs (purple, anti-ChAT), which lie at or very near the same level as the R-cell ON processes. F, Three closely spaced R-cells (red; white arrows) with visible axons, labeled in the Rbp4-Cre-tdTomato mouse. The other red cells are Rbp4-Cre-labeled displaced ACs (RACs). Shown for comparison are the somas and axons of ipRGCs (cyan, anti-melanopsin) and the somas of ON starburst ACs (gold, anti-ChAT). G, The three closely spaced R-cells show a hint of mosaics formed by the dendritic arbors in both the ON (green) and OFF IPL sublayers (purple). Maximum intensity projection of several optical sections centered on the relevant arbor.

R-cells have sustained ON irradiance-encoding light-evoked responses

R-cells exhibited sustained ON responses to whole-field illumination, and only a brief hyperpolarization at OFF (Fig. 7A). Higher light intensities resulted in reduced spike amplitude and regularity (Fig. 7A), presumably due to partial sodium-channel inactivation. The average firing rate (Fig. 7B) and the IR function derived from it (Fig. 7C) revealed only modest dependence of spiking on stimulus intensity. However, when clamping the membrane at the chloride reversal potential, the irradiance-dependence of the steady-state, light-evoked inward current became more apparent (Fig. 7D,E). Inward currents at light offset were also observed (Fig. 7D). The magnitudes of these, too, were irradiance dependent.

• Download figure
• Open in new tab
• Download powerpoint

Figure 7.

Rbp4 R-cells show sustained irradiance-encoding light-evoked responses. A, Light-evoked firing, obtained in whole-cell current-clamp configuration, to a series of stimulus intensities (irradiance) under control conditions (retina bathed in Ames’ medium only). Irradiances in log photons cm⁻² s ^{− 1} are indicated above traces. Horizontal black bar below the leftmost trace indicates the duration of a 15 s wide-field light step from darkness. B, PSTHs averaged across cells (n = 4 cells; average ± SEM) for various irradiance levels. C, IR curve based on data presented in B. Data points and error bars indicate the steady state response (average ± SEM). The coefficient of determination (R ²) of the fit is indicated. D, Light-evoked currents, obtained in whole-cell voltage-clamp configuration, to a series of stimulus irradiance levels under control conditions. The membrane voltage was clamped at the chloride reversal potential. E, IR curve based on data presented in D. F–N, Pharmacological analysis of circuitry underlying R-cell light responses. F–H, RGC responses, obtained in whole-cell current-clamp configuration, while blocking the ON pathway using L-AP4. F, Example of a cell’s light-evoked voltage response (black trace) to the brightest stimulus tested (13 log photons cm⁻² s ^{− 1}). The lower envelope of the voltage trace is shown in magenta. Inset, Same trace at higher gain showing a sustained hyperpolarization in the voltage response as well as the lower voltage envelope throughout the stimulus duration. G, Voltage envelope (n = 3 cells; average ± SEM) for various irradiance levels. H, IR curve based on data presented in G. I–K, RGC responses (n = 5 cells), obtained in whole-cell current-clamp configuration, while blocking both the ON and OFF pathways using L-AP4, D-AP5, and DNQX. Conventions in individual plots are the same as in F–H. L–N, R-cell responses (n = 4 cells), obtained in whole-cell current-clamp configuration, after more complete synaptic blockade by further addition of the OFF channel blocker ACET to the L-AP4, D-AP5, and DNQX already in the bath. Conventions in individual plots are the same as in F–H. O, Firing rate in response to a series of bright spots on a dark background at varying sizes (main plot, 50-400 μm; inset, 50-1200 μm; n = 3 cells). P, Example of direction and orientation selectivity tuning of two R-cells. Thick blue lines show the average cells’ response to sinusoidal gratings drifting in each of eight different directions, 45° apart. Thin blue lines show data from each of four single trials. N, nasal; D, dorsal; T, temporal; V, ventral. Red lines indicate the preferred direction based on four individual repetitions (thin lines), and their average (thick line). The DSI, OSI, and preferred direction (PD) of the cell are indicated. Q, Average ± SEM DSI and OSI across tested cells (n = 5 cells).

Blocking the ON channel with L-AP4 reversed the sign of the steady-state light response from depolarization and spiking (Fig. 7A) to hyperpolarization (Fig. 7F). Although the hyperpolarization was relatively modest, it was accompanied by a reduction in membrane noise, and a dramatic suppression of spiking (Fig. 7F). Both onset and offset of the light stimulus evoked a small, transient hyperpolarization. Membrane voltage varied only modestly as a function of stimulus irradiance (Fig. 7G,H).

We next supplemented the ON-channel blockade with agents designed to silence the OFF pathway (namely, a mix of NMDA and AMPA/kainate blockers, D-AP5 and DNQX). Although this strongly attenuated the transient hyperpolarizations at light ON and OFF, we were surprised to find that this incompletely blocked the sustained ON hyperpolarization, and that light offset induced poststimulus bursting in several cells (Fig. 7I). Membrane voltage was minimally related to stimulus irradiance under these conditions (Fig. 7J,K). We suspected that the residual light responses were due to incomplete kainate receptor blockade, so we added ACET (a selective antagonist for GluR5-containing kainate receptors). This rendered the cells largely irresponsive to light (Fig. 7L–N). Taken together, the data suggest that ON bipolar input normally dominates in R-cells, but that they also receive OFF bipolar input that is unmasked when the ON channel is silenced.

To study center-surround organization, we presented bright spots of various sizes (Fig. 7O). On average, the optimal spot diameter was 280 μm, which is substantially larger than the dendritic arbor. Larger spots attenuated the response, indicating the presence of an inhibitory surround (Fig. 7O, inset). R-cells generally showed some tuning for the direction or orientation of drifting gratings, but such tuning was inconsistent and typically weak (Fig. 7P,Q; DSI = 0.21 ± 0.13; OSI = 0.36 ± 0.22).

Serial EM reconstruction of an R-cell and its synaptic inputs

To characterize the types of BCs that synapse onto R-cells, we mined a SBEM volume (k0725) generated from a wild-type adult mouse retina (Ding et al., 2016). The volume (50 × 210 × 260 μm³) spanned the IPL and the immediately adjacent portions of the GCL and INL. The retina had been processed so as to reveal intracellular structures such as synaptic ribbons and vesicles.

Among many RGCs traced in the EM volume, we identified diverse bistratified forms. One of these RGCs closely resembled the R-cells as just characterized at the light microscopic level (compare Figs. 2B, 8A ). Several key morphologic measures obtained from the EM-reconstructed presumptive R-cell fell within 1 SD of the mean value of those measures for R-cells traced in confocal z-stacks (Table 1, compare R-cell and R-cell EM). The dendritic tree of the EM-reconstructed cell were similar in form to those of R-cells in branching structure, apparent dendritic field size and stratification (Fig. 8A). The outer arbor of the reconstructed R-cell stratified between the dendrites of reconstructed M1 ipRGCs and the OFF ChAT band (inferred from dendritic stratification of reconstructed ON-OFF DS cells). The inner arbor resided within and just below the ON ChAT band (Figs. 6D, 8B ). Although the soma diameter of this presumptive R-cell was larger than those of R-cells as assessed by confocal microscopy (16.1 vs 12.9 ± 1.0 μm), this might have resulted from osmotic factors during tissue fixation (Thorne and Nicholson, 2006; Mikula and Denk, 2015).

• Download figure
• Open in new tab
• Download powerpoint

Figure 8.

Serial EM reconstruction of an Rbp4 RGC and its synaptic inputs. A, EM reconstruction of a presumptive Rbp4 R-cell (black) along with dyad synapses made onto it (red dots), as viewed en face (top) and in an orthogonal plane (bottom). Scale bar, 50 μm applies to panels A--D. B, Orthogonal view of the presumptive R-cell (black) along with traces of several ON-OFF DS RGCs (yellow) and two M1 ipRGCs (red). The outer dendritic arbor of the presumptive R-cell stratifies below the M1 ipRGC plexus and just above the OFF ChAT band; its inner arbor stratifies just below the ON ChAT band. C, Orthogonal view of BCs making identified ribbon synaptic contacts onto the presumptive R-cell. Top and bottom gray horizontal bars represent the OFF and ON ChAT bands, respectively. D, EM reconstruction of a presumptive R-cell (black) along with BCs that synapse onto it, as viewed en face (top) and in an orthogonal plane (bottom). BC colors follow the scheme indicated in the key below. Note that the presumptive R-cell did not receive input from rod BCs (RB) or cone BCs of types 4 and 9. E, Percentage of all bipolar input to the presumptive R-cell as a function of BC type. OFF input derived mainly through types 1 and 2, and ON input derived mainly through type 7. F, Axon terminal field area (median and first and third quartiles) as a function of BC type. G, IPL depth (median and first and third quartiles) as a function of BC type. H, Mean normalized density of axonal processes as a function depth for each BC type. Although RB cells lacked synaptic contacts onto the presumptive R-cell, they were included in the analysis as a useful benchmark in the assessment of depth of axon-arbor stratification for BC types (see Materials and Methods).

To identify the BC types that synapse onto the presumptive R-cell, we scanned its dendrites for ribbon synaptic contacts (Fig. 8A, red dots); subsequently, starting from randomly selected identified dyad synapses (128 out of 281), we reconstructed the complete axon and axon terminal of 76 BCs. To identify BC types, we assessed the stratification of their arbors in relation to one another and to the cholinergic bands (Fig. 8C,G,H). Figure 8D shows all reconstructed BCs synapsing onto the presumptive R-cell as projected onto the plane of the retina (D, top; en face view) or onto an orthogonal, vertical plane to show depth information (bottom). ON synaptic input to the presumptive R-cell was dominated by type 7 ON cone BCs, while its OFF synaptic input derived almost entirely from type 1 and 2 OFF cone BCs (Fig. 8D,E). Figure 8F–H quantify two features of axon terminals that distinguish bipolar types from one another, the dimensions of the terminal arbor (Fig. 8F) and their depth within the IPL (Fig. 8G,H). In general, the presumptive R-cell seemed to receive input from axon terminals of all BC types within its dendritic field, provided they costratified with the R-cell’s dendrites. ON inputs were more common than OFF inputs (78 vs 50 dyads in the inner and outer arbors; 53 vs 25 ON and OFF cone BCs), paralleling the dominant ON response observed physiologically (Fig. 7A–E).

A novel BC type, cone BC type 0

Among the presynaptic neurons providing ribbon synaptic contact onto the presumptive R-cell, we identified four cells that appear to belong to a previously unknown BC type (Fig. 9). They superficially resembled types 1 and 2 OFF cone BCs but were clearly different from both. Most of the type 0 cells (three of the four) had cell bodies at the inner margin of the INL (the remaining cell’s soma lay outside the volume). This is a highly atypical position for BC bodies, but is characteristic of ACs. However, there is no precedent for synaptic ribbons in ACs, so for now we assume that this is a novel OFF cone BC type, which we provisionally term “type 0.”

• Download figure
• Open in new tab
• Download powerpoint

Figure 9.

A novel OFF BC, type 0. A–C, Partial mosaics of cone BC types 0, 1, and 2, as viewed en face (top) and in an orthogonal plane (bottom). Top and bottom gray horizontal bars represent the OFF and ON ChAT bands, respectively. Scale bar, 50 μm. The soma of one type 0 was not included in the SBEM volume, perhaps due to slight variation in the orientation of the retina within the SBEM volume. D, Terminal field area (median and first and third quartiles) of type 0 BCs was larger than those of types 1 and 2. E, IPL depth (median and first and third quartiles) of type 0 cells was larger than those of types 1 and 2. Gray horizontal bar represents the OFF ChAT band. F, Mean normalized density of axonal processes of type 0 BCs was slightly more peaked and narrow than types 1 and 2. Vertical dashed line represents the OFF ChAT band. G, Type 0 BCs made both dyad and monad synapses; arrowheads mark ribbons. All contacts onto the presumptive R-cell were dyads; the postsynaptic partners of monads are unknown.

Multiple features differentiated type 0 from type 1 (n = 25) or type 2 (n = 16) BCs. First, the arbors of type 0 BCs were sparsely branched and significantly broader than those of types 1 and 2 (Fig. 9D; 1614 μm², 450 μm², and 618 μm² axon terminal field areas for types 0, 1, and 2; permutation one-way ANOVA, p < 0.001; by post hoc pairwise comparisons, arbor areas of type 0 were significantly larger than those of type 1 or 2, p < 0.001). Second, their axonal arbors stratified slightly more proximally (lower) in the IPL [Fig. 9E,F; IPL depth of 16%, 12%, and 9% for types 0, 1, and 2; permutation one-way ANOVA, p > 0.05; post hoc pairwise comparisons, type 0 axons arborized significantly deeper in the IPL than type 2 axons (p = 0.048), whereas the depth difference with type 1 axons did not reach statistical significance (p = 0.173)]. Finally, their arbors did not fit into the mosaics of BC type 1 and 2 axon terminal fields, and instead dovetailed with each other in a way that suggests an independent tiled mosaic in the IPL (Fig. 9A–C). Like other BCs, type 0 BCs make mainly dyadic ribbon contacts in the IPL.

Morphology and physiology of RDS cells

The other RGC type labeled in the Rbp4-Cre line, the DS cell here termed RDS, was rarer than R-cells, and easily distinguishable from them by virtue of its much larger soma (RDS: 17.5 ± 2 μm, n = 4; R-cells: 12.9 ± 1.0 μm, n = 16). RDS cells also had larger dendritic arbors. This was because the OFF arbor was larger in RDS cells (diameter, 194 ± 27 μm; n = 4) than in R-cells (138 ± 36 μm; n = 13); in contrast, ON arbor dimensions were very similar between the two types (RDS diameter, 177 ± 15 μm; R-cell diameter, 178 ± 35 μm). In branching structure and stratification, RDS cells strongly resembled other ON-OFF DS types (Figs. 2C, 10A; Huberman et al., 2009; Rivlin-Etzion et al., 2011; Trenholm et al., 2011; Dhande et al., 2013). Like those types, the RDS cell restricted its dendritic arbors to the two sublaminae containing processes of starburst ACs, revealed by ChAT immunolabeling (Fig. 10B). Moreover, like other ON-OFF DS cells, the RDS cell’s dendrites cofasciculated with the ChAT-immunopositive processes (Fig. 10C) and avoided the melanopsin immunoreactive plexuses at the margins of the IPL (Fig. 10B). We confirmed the direction selectivity of RDS cells by showing that they responded differentially to the direction of motion of sinusoidal gratings and bars. RDS cells were ON-OFF cells, responding to both the leading and trailing edge of the moving bar (Fig. 10E). Although our sample is small, all recorded RDS cells preferred ventral motion on the retina (Fig. 10E,F). They are thus reminiscent of cells labeled in the Hb9-GFP mouse line, which are also ON-OFF DS cells that prefer ventral retinal motion (Trenholm et al., 2011). Like Hb9-GFP cells, RDS cells exhibited dendritic asymmetry (Fig. 10A), with the center of the dendritic arbor displaced from the cell body in the cell’s preferred direction (compare Fig. 10F,G).

• Download figure
• Open in new tab
• Download powerpoint

Figure 10.

Morphology of DS RGCs in the Rbp4-Cre mouse. A, Drawings of three DS RGCs labeled in the Rbp4-Cre mouse (RDS cells). B, Dendritic stratification of an RDS cell. Five views of the same rotated orthographic maximum intensity projection derived from z-stacks taken in the vicinity of a single labeled RDS cell are shown. Views (top to bottom) show the individual tdTomato in the Rbp4:Cre;Ai14 mouse (tdT), tdTomato and anti-ChAT labeling (tdT/ChAT), tdTomato and anti-melanopsin labeling (tdT/mel), anti-ChAT and anti-melanopsin labeling (mel/ChAT), and tdT/mel/ChAT labeling. RDS dendrites with Rbp4-Cre-dependent viral labeling (red) match the depth of the ON and OFF ChAT bands [ON and OFF starburst AC (SAC) arbors, blue]. Both Rbp4 (red) and ChAT (green) labeling appear below the outer M1 ipRGC plexus and above the inner M2 ipRGC plexus (purple). C, Dendrites of the inner arbor of a virally labeled RDS cell (green) cofasciculate with the processes of ON starburst ACs (anti-ChAT) which lie at or very near the same levels as the RDS-cell inner processes. Scale bar, 50 μm. D, Voltage responses of a representative RDS cell to sinusoidal contrast gratings drifting in eight directions at 45° intervals. Polar plot shows response amplitude as a function of stimulus direction (normalized to maximum response). Red vector shows preferred direction in retinal coordinates (N, nasal; D, dorsal; T, temporal; V, ventral). E, Voltage traces for the same representative RDS cell, illustrating the ON and OFF responses to the leading and trailing edges of a drifting bright bar. F, A polar plot summarizing the direction preference (vector angle) and DSI (vector length). RDS cells preferred ventral motion on the retina. G, A polar plot summarizing the direction of dendritic arbor asymmetry (vector angle) and asymmetry index (AI, vector length), for the inner (blue) and outer (green) arbors. H, I, Both types of Rbp4-cre RGCs (R-cells and RDS cells) are CART immunoreactive. Triple labeling in the GCL for Rbp4-Cre-tdTomato (red), CART (green; an immunomarker for ON-OFF DS RGCs), and RBPMS (magenta; an immunomarker for RGCs). Cre-positive ganglion cells are indicated by white markers. Most are R-cells (hollow white arrowheads), but two are RDS cells (filled white arrows). All of these are also immunopositive for RBPMS and CART (I). Cre-positive displaced ACs (RACs), indicated by blue markers, are identifiable from their RBPMS immunonegativity (I). Virtually all of these lack CART immunolabeling (hollow blue arrowheads), but one RAC appeared to be CART-immunopositive (solid blue arrow). Scale bar, 50 μm.

Like most ON-OFF DS types (Kay et al., 2011; Dhande et al., 2013), RDS cells were clearly immunoreactive for CART (Fig. 10H,I, white arrows). To our surprise, the R-cell, the other RGC type labeled in this line, was also virtually always CART-immunopositive despite its weak directional tuning (Fig. 10H,I, hollow white arrowheads). In the GCL, 97% of all RGCs (identified by their RBPMS labeling) that exhibited Cre-driven labeling in the Rbp4-Cre mouse also expressed CART (n = 64), whereas all but a handful of Cre-labeled displaced ACs (RACs; Fig. 10H,I, hollow blue arrowheads) were CART-negative. Two identified displaced Rbp4-Cre-labeled RGCs also lacked CART staining.

Axonal projections of Rbp4-Cre RGCs

To study the central projections of RGCs in the Rbp4-Cre line, we made intraocular injections of Cre-dependent mCherry viruses. The advantage of this viral approach, as compared with crossing Rbp4-Cre mice with Cre reporter mice, is that it ensured that all fluorescent axons were of retinal origin. Labeled axon terminal fields were prominent in the dorsal devision of the lateral geniculate nucleus (dLGN). There, they were concentrated within a specific lamina, at roughly the transition between the external shell and main core of the nucleus (Fig. 11A,C), defined by their differential input from DS or melanopsin-expressing RGCs, respectively (Huberman et al., 2009; Ecker et al., 2010; Kay et al., 2011; Rivlin-Etzion et al., 2011). The superior colliculus (SC) was also prominently labeled, with highly branched fine axonal arbors that terminated only very superficially in the upper retinorecipient layers (Fig. 11D). Additional terminal fields were detected in a thin sheet at the lateral margin of the ventral LGN (Fig. 11A,C), and in the nucleus of the optic tract (NOT), a component of the accessory optic system (Fig. 11A–C). No labeling was detected in any other targets of retinofugal axons, including the medial terminal nucleus (MTN) and suprachiasmatic nucleus (SCN).

• Download figure
• Open in new tab
• Download powerpoint

Figure 11.

Brain projections of Rbp4 RGCs, and R-cells alone. A–D, Axon terminal labeling following injection of a Cre-dependent viral reporter. A, Labeling was evident in the LGN (A, C) and NOT (A, B). C, Two coronal sections, consecutive along the rostral-caudal axis. Labeling was evident in the dorsal and ventral divisions of the LGN (dLGN, vLGN), but not in the intergeniculate leaflet (IGL). Labeling was restricted to a narrow band along the lateral margin of the vLGN. D, Intense labeling was evident also in the superficial layers of the SC. E, Axon terminal labeling following sparse labeling of Cre-expressing neurons. Only three RGCs were detectably labeled in this retina, and all were bistratified R-cells. Axon terminal labeling was detected only in the dLGN (at the boundary between core and shell regions), in the vLGN (left), and in the most superficial layers of the SC (right).

These axon terminal fields represent the combined projections of R-cells and RDS-cells, because the intraocular virus used transduced both RGC types, judging from their somadendritic architecture. However, in one case, viral transfection was sparse enough to label only three RGCs across the whole retina. All three were clearly R-cells. Their three axons were easily traced to the optic nerve head and into the brain. Three distinct arbors were evident in the dLGN, vLGN, and SC (Fig. 11E), but no axon terminals were detected in the NOT. These data suggest that R-cells innervate the LGN complex and SC, but not other retinorecipient targets, including the accessory optic system, while the RDS cells innervate at least the NOT.