Receptive field properties of bipolar cell axon terminals in direction-selective sublaminas of the mouse retina

Abstract

Retinal bipolar cells (BCs) transmit visual signals in parallel channels from the outer to the inner retina, where they provide glutamatergic inputs to specific networks of amacrine and ganglion cells. Intricate network computation at BC axon terminals has been proposed as a mechanism for complex network computation, such as direction selectivity, but direct knowledge of the receptive field property and the synaptic connectivity of the axon terminals of various BC types is required in order to understand the role of axonal computation by BCs. The present study tested the essential assumptions of the presynaptic model of direction selectivity at axon terminals of three functionally distinct BC types that ramify in the direction-selective strata of the mouse retina. Results from two-photon Ca²⁺ imaging, optogenetic stimulation, and dual patch-clamp recording demonstrated that 1) CB5 cells do not receive fast GABAergic synaptic feedback from starburst amacrine cells (SACs); 2) light-evoked and spontaneous Ca²⁺ responses are well coordinated among various local regions of CB5 axon terminals; 3) CB5 axon terminals are not directionally selective; 4) CB5 cells consist of two novel functional subtypes with distinct receptive field structures; 5) CB7 cells provide direct excitatory synaptic inputs to, but receive no direct GABAergic synaptic feedback from, SACs; and 6) CB7 axon terminals are not directionally selective, either. These findings help to simplify models of direction selectivity by ruling out complex computation at BC terminals. They also show that CB5 comprises two functional subclasses of BCs.

bipolar cells (BCs) are interneurons in the vertebrate retina that transmit visual signals from photoreceptor cells to the output neurons (ganglion cells). In the mammalian retina, 10–12 types of cone BCs have been identified, each with a characteristic axonal stratification pattern within the inner plexiform layer (IPL) (Ghosh et al. 2004; Helmstaedter et al. 2013; Light et al. 2012; MacNeil et al. 2004; Masland 2012; Wassle et al. 2009). BC types decompose visual information into parallel channels and feed their signals to different strata of the IPL, where complex synaptic computations are performed, allowing different ganglion cell types to extract different features of the visual world and relay them to the brain along parallel visual pathways (Masland 2012; Pang et al. 2004; Roska and Werblin 2001). The functional properties of BCs are thought to be cell type specific and correlated closely with the axonal stratification pattern (Masland 2012). The primary segregation of cone BCs is into the OFF and ON layers of the IPL, corresponding to sublaminas a and b, respectively (Famiglietti et al. 1977; Famiglietti and Kolb 1976; Nelson et al. 1978). Within this primary classification, BC types that stratify near the middle of the IPL give more transient light responses, whereas those that stratify near IPL borders have more sustained responses (Baden et al. 2013; Borghuis et al. 2013). However, the spatial receptive field properties of BC types are not well understood, especially at the level of axon terminals (i.e., their contribution to ganglion cell computations).

An interesting question about BC receptive field properties relates to the possibility that some BC axon terminals may serve as a substrate for complex network computation. In particular, it has been proposed, based on the spatially asymmetric glutamate currents recorded from direction-selective ganglion cells (DSGCs) under patch clamp (Borg-Graham 2001; Fried et al. 2002, 2005; Lee et al. 2010; Sun et al. 2006; Taylor and Vaney 2002; Weng et al. 2005), that certain BC axon terminals may be directionally selective, possibly as a result of spatially asymmetric GABAergic feedback inhibition from starburst amacrine cells (SACs) (Fried et al. 2002, 2005). In this model, there are three parsimonious assumptions that must be tested. First, SACs make GABAergic feedback synapses onto BC terminals that are presynaptic to DSGCs. Previous electron microscopy (EM) studies have found no evidence for such feedback synapses in the rabbit retina (Famiglietti 1991), but reciprocal synapses have been reported between choline acetyltransferase (ChAT)-immunoreactive processes and a subpopulation of BC axon terminals in the primate retina (Yamada et al. 2003). It is important to determine whether SACs make functional inhibitory synapses onto the BC axon terminals that contact DSGCs and SACs. Second, because the number of DSGC types is thought to exceed that of the BC types providing their input (Vaney et al. 2012), some local regions of the BC axon terminal are expected to function autonomously, so that they can selectively synapse onto multiple types of DSGCs with different preferred directions. This type of compartmentalized axonal computation would mirror the dendritic computation occurring in the postsynaptic SAC dendrites (Euler et al. 2002; Fried et al. 2002, 2005; Hausselt et al. 2007; Lee et al. 2010; Lee and Zhou 2006; Wei and Feller 2011; Yonehara et al. 2011) and would be consistent with a recent study that shows that a single BC can send different signals to different ganglion cells (Asari and Meister 2012). However, because BC terminals that ramify in the SAC and DSGC strata are anatomically compact, it must be tested directly whether these terminals actually have the electrotonic structure to support compartmentalized computation. Third, BC terminals that synapse on DSGCs are expected to show direction-selective tuning of their light responses. According to a recent report (Yonehara et al. 2013), BCs that are transsynaptically infected by virus from the ON type of DSGCs do not give directionally selective Ca⁺² responses to light stimulus movement. However, it remains to be determined whether this result also applies to ON-OFF DSGCs, which may receive input from CB7 cells (Shi et al. 2011) in addition to CB5 cells (but see Helmstaedter et al. 2013) or from a subpopulation of CB5 cells that might have failed to be infected transsynaptically from ON DSGCs (CB5 and CB7 cells are also named Type 5 and Type 7 cells, respectively; Ghosh et al. 2004).

Another interesting question regarding BC receptive field properties is whether BCs that stratify at a similar depth within the IPL all share similar receptive field properties. The current classification of BC types in the mouse retina relies heavily on axonal morphology and stratification depths in the IPL (Ghosh et al. 2004; Wassle et al. 2009). It is often believed that each morphological BC type corresponds to a functional type. However, morphologically similar BCs may comprise subtypes that are different in synaptic connectivity (Helmstaedter et al. 2013). On the basis of dendritic density and cone-to-bipolar ratio, morphologically identical CB5 cells in the mouse retina are suggested to contain two subtypes, CB5a and CB5b (Wassle et al. 2009). It has been unclear whether morphologically similar CB5 cells actually consist of functional subtypes with different receptive field properties, including direction-selective properties.

This study investigated the functional properties of individual BC axon terminals that ramify near the ON cholinergic stratum (sublamina 3/4) of the mouse retina, focusing on the above two interesting questions. Using a combination of in vivo genetic transfection, two-photon imaging, optogenetics, and dual patch clamp in the wholemount mouse retina, we demonstrated that CB5 cells did not receive fast GABAergic synaptic feedback from SACs, that Ca²⁺ signals in various parts of CB5 axon terminals were well correlated, and that CB5 axon terminals did not show direction selectivity to moving light stimulation. Notably, we discovered that morphologically similar CB5 cells had two distinctly different receptive field structures, suggesting the existence of at least two functional subtypes, but neither subtype was directionally selective. We also found that CB7 cells provided strong inputs to ON SACs but did not receive direct GABAergic synaptic feedback from SACs, and that CB7 axon terminal branches did not show direction selectivity in their Ca²⁺ responses to moving light stimulation.

METHODS

Mouse lines, in vivo electroporation of retina, and retinal wholemount preparation.

All procedures involving the use of animals were performed in accordance with National Institutes of Health guidelines and protocols reviewed and approved by the Yale University Animal Care and Use Committee. ChAT-Cre [strain B6;129S6-Chat^tm1(cre)Lowl/J], tdTomato [strain B6.Cg-Gt(ROSA)26Sor^{tm9(CAG-tdTomato)Hze}/J], ChR2-YFP [strain B6;129S-Gt(ROSA)26Sor^{tm32(CAG-COP4*H134R/EYFP)Hze}/J], and wild-type (strain C57BL/6J) mouse lines were obtained from The Jackson Laboratory (Bar Harbor, ME). Crossing the ChAT-Cre line with the tdTomato line and with the ChR2-YFP line produced offspring (ChAT-Cre/tdTomato and ChAT-Cre/ChR2-YFP) that expressed, respectively, tdTomato and ChR2-YFP in cholinergic amacrine cells in the retina.

Plasmid DNA encoding GCaMP3 under the mGluR6 promoter was prepared as previously described (Tian et al. 2009). In vivo electroporation was performed to transfect retinal progenitors at P1–P3 (Matsuda and Cepko 2004). Newborn mouse pups were anesthetized by chilling on ice, and the eyelids were opened. A plasmid DNA solution in 10 mM Tris·HCl (0.2 μl at 4.8 μg/μl) was injected into the subretinal space via a sharp glass pipette with a microinjection system (PLI-100 Pico Injector, Harvard Apparatus) under a dissection microscope. Alternatively, after the eyelids were opened under anesthesia, a small incision was made in the sclera with a 30-gauge needle and the plasmid DNA solution was injected through the incision with a handheld Hamilton syringe with a 33-gauge blunt-ended needle. Electroporation was performed by applying five square-wave electrical pluses (80 V, 50-ms duration, 950-ms interval) to the eyes with an electroporation system (ECM 830, Harvard Apparatus) and homemade tweezer-type electrodes. In vivo transfection of GCaMP3 in retinal ganglion cells was achieved by intraocular injection of an AAV2/2.synapsin1.GCaMP3.

To prepare wholemount retinas, 3- to 8-wk-old mice of either sex were killed by an injection of Euthasol (>2 ml/kg ip) after 1–3 h of dark adaptation. The eyes were quickly enucleated and hemisected. The retinas were isolated from the retinal pigment epithelium under a dissection microscope in ACSF (composition in mM: 125 NaCl, 2.5 KCl, 2 CaCl₂, 1 MgCl₂, 1.25 NaH₂PO₄, 0.5 l-glutamine, 26 NaHCO₃, and 20 d-glucose) equilibrated with 95% O₂-5% CO₂ at room temperature (20–25°C). A piece of retina was placed in a recording chamber and held down to the chamber bottom, ganglion cell side facing up, by a platinum tissue holder. The recording chamber was placed on the stage of a two-photon microscope and perfused continuously with ACSF at ∼32°C at a rate of ∼4 ml/min. The entire tissue preparation process was completed under dim red light illumination.

Two-photon imaging and visual stimulation.

Fluorescence measurements were made in a lightproof Faraday cage with a two-photon microscope system (Ultima, Prairie Technologies, Middleton, WI) configured on an Olympus upright microscope (BX51WI, Olympus USA) with a ×60, 1.0 NA objective (LUMPlanFL/IR, Olympus) and a Ti:sapphire pulsed laser (MaiTai, Newport, CA) tuned to 910–920 nm. Image acquisition was controlled by Prairie software (Prairie View, Prairie Technologies). Fluorescence signals were detected simultaneously by two photomultiplier tubes at 520 ± 18-nm (green channel, band-pass filter: Semrock, FF01-520/35-25) and 607 ± 23-nm (red channel, band-pass filter: Chroma, HQ607/45) wavelength, respectively. Fluorescence signals from small regions (∼50 × 50 μm²) of the image field containing GCaMP3-expressing axon terminals were recorded in the green channel at a rate of 10–30 frames/s (scanning laser dwell time: 4 μs/pixel). The timing of the visual stimuli was recorded simultaneously in the red channel. z-Series fluorescence images were collected at a z-step size of 0.5–1 μm in both green and red channels. Three-dimensional (3D) rendering of the images was performed with the software ImageJ (http://rsb.info.nih.gov/ij) to discern the morphology of and the spatial relationship between GCaMP3-expressing BCs and tdTomato-expressing cholinergic processes and/or rhodamine 101-filled DSGC dendrites in the IPL.

Visual stimuli were generated with Vision Works software (Vision Research Graphics, Durham, NH) and presented by a miniature transmissive TFT LCD (model XGAP01, CRL OPTO, Dalgety Bay, UK; 36.9 mm × 27.6 mm in dimension, 1,024 × 768 pixels, contrast ratio > 1:100), which was transilluminated by collimated light from a halogen lamp, band-pass filtered at 577 ± 10 nm (Semrock, FF01-577/20). The LCD image was projected to the photoreceptor outer segment layer of the retina through the microscope condenser lens. The maximum intensity of the LCD image at the retina was 3.4 × 10⁻¹³ W/μm² (measured with a radiometer, model S470, UDT Instruments, San Diego, CA). The light stimulus, consisting of either a stationary light spot of variable diameter (30–1,400 μm) or a light bar (100 μm high, 500 μm long) moving across the receptive center in four orthogonal directions at 500 μm/s, was presented on the retina at an intensity of 3.4 × 10⁻¹³ W/μm² over a uniform background light of 3.4 × 10⁻¹⁴ W/μm². The constant background light helped the retina adapt to the scanning IR laser light used for two-photon imaging.

Patch-clamp recording and photoactivation of channelrhodopsin-2.

Single and dual patch-clamp recordings were made from ON cone BCs and ON-OFF DSGCs in wholemount retinas of ChAT-Cre/ChR2-YFP mice with methods similar to those previously described (Lee et al. 2010; Zhou 1998). Some patch-clamp experiments were also repeated in mouse retinal slices (200 μm thick) with methods modified from those previously described (Zhou et al. 1994; Zhou and Fain 1995). The whole cell pipette solution contained (in mM) 105 CsMeSO₄, 0.5 CaCl₂, 10 HEPES, 5 EGTA, 5 Na₂-phosphocreatine, 2 ATP-Mg, 0.5 GTP-2Na, 2 ascorbic acid, and 8 QX314-Cl, titrated to pH 7.2 with CsOH. Liquid junction potential was calculated with pCLAMP software and corrected (Molecular Devices, Union City, CA). Alexa Fluor 594 (0.28 mM) was included in the whole cell pipette solution. The morphology of patch-clamped cells was identified at the end of the recording by taking two-photon z-series images of the recorded BCs (red channel), which were reconstructed in 3D in ImageJ and compared with the location of ChR2-YFP-expressing cholinergic processes in the IPL (green channel). Pharmacological drugs were applied to the retina via bath perfusion.

ChR2 was activated with light from a 100-W Hg bulb (from the epifluorescence port of the microscope), filtered by a band-pass filter at 465 ± 15 nm and controlled by a shutter (Uniblitz, Vincent Associates, Rochester, NY). The intensity of this blue light was 8 nW/μm² (2 × 10¹⁰ photons·s⁻¹·μm⁻²) measured at the retina.

Image processing and data analysis.

Fluorescence responses from GCaMP3-expressing BCs were analyzed with ImageJ and custom-written scripts in MATLAB (MathWorks, Natick, MA). x–y sample movement in image series during long recordings was corrected with a template matching method using a plugin in ImageJ (https://sites.google.com/site/qingzongtseng/template-matching-ij-plugin). Regions of interest (ROIs) were drawn manually around the contours of fluorescent segments of BC terminals. Fluorescence responses were averaged spatially over each ROI. After background subtraction (intensity measured from a ROI subtracted by the intensity measured from a region containing no GCaMP3-expressing processes), the fluorescence response (R) of each ROI was calculated as a fractional response as follows:

$$\text{R}=\Delta \text{F}/\text{F}=\frac{\text{F}-{\text{F}}_0}{{\text{F}}_0}$$

where F is the background-subtracted fluorescence intensity of a light-evoked response (or a spontaneous event); F_o is the background-subtracted baseline intensity averaged over a 1-s period immediately prior to each visual stimulation (or spontaneous event). In the case in which the response to the onset of the scanning IR laser was measured (see Fig. 3E), F_o was defined as the background-subtracted, steady-state intensity measured after the IR laser response had become adapted during continuous scanning. For statistical comparison, light responses to repeated (4–6) trials of the same stimulation were averaged and mean peak response amplitudes were used. A directionally selective index (DSI) was calculated as

$$\text{DSI}=\frac{\sqrt{{\left({\text{R}}_0-{\text{R}}_{180}\right)}^2+{\left({\text{R}}_{90}-{\text{R}}_{270}\right)}^2}}{{\text{R}}_0+{\text{R}}_{90}+{\text{R}}_{180}+{\text{R}}_{270}}$$

Fig. 3.

Comparison of response amplitude and kinetics among different regions of CB5 axon terminals. A: cross-sectional view of reconstructed GCaMP3-expressing BCs (green) in a ChAT-Cre/tdTomato mouse retina, showing 2 CB5 cells (cell 1 and cell 2, with their axon terminals boxed in white and blue rectangles, respectively), a putative CB8 cell (cell 3, with axon terminal boxed in red rectangle), rod BC axons (white arrowheads), and SAC somas and processes (red). B: tangential 2-photon image of axon terminal branches of cell 1 (green/yellow) and cell 2 (green/yellow in blue box) from A, showing cofasciculation with ON SAC processes (red). The axon of the CB8 in A appeared in this image as a single dot (cell 3, in red box). C: same image as in B, showing only GCaMP3 signals (green channel), with ROIs drawn over various regions of the 2 CB5 axon terminals (white and blue, respectively) and the CB8 axon stalk (red). D, bottom: overlay of fluorescence signals (ΔF/F) measured as a function of time from all ROIs (1–5 on cell 1 and 1 and 2 on cell 2), over the 2 CB5 terminals in C showing spontaneous activities (*) and responses to flashes (shaded yellow) of an 80-μm-diameter light spot at 4 different positions (pos 1–4) in the receptive field. Inset, positions of the light spot relative to the receptive field (RF) center (green circle). Top: snapshot images of spontaneous and light-evoked Ca²⁺ signals (in pseudocolor) in CB5 terminal branches at indicated time points. Same view as in B. E: expanded view of the Ca²⁺ responses from different regions of the two CB5 cells (cell 1: ROIs 1–6, black traces; cell 2: ROIs 1 and 2, blue traces), showing similar response kinetics from all ROIs for both spontaneous and light-evoked responses. cell 1 Norm and cell 2 Norm: overlay of responses from all ROIs on cell 1 and cell 2, respectively, after normalization of the response amplitudes by the peak amplitudes of responses to the initial IR scanning laser, showing nearly identical response kinetics and relative amplitude from all ROIs in CB5 cells, which were very different from the response from the axon stalk of the CB8 cell (red trace). F and G: 2-photon image from various ROIs of a CB5 axon terminal in a different retina (F), showing overlay of light-evoked responses with similar kinetics and relative response amplitudes (G). On rare occasions, spontaneous activities (*) were observed only locally in 1 (ROI 4, see inset), but not the rest, of the ROIs. Right: magnification of boxed area on left. Scale bars in all images, 10 μm.

where R₀, R₉₀, R₁₈₀, and R₂₇₀ correspond to the response amplitude to stimulus moving at 0°, 90°, 180°, and 270°, respectively. Statistical significance was determined at the level of P = 0.05 by paired-sample t-test between the responses to a stimulus moving in opposite directions, or by one-way ANOVA among DSIs of different populations of axon terminals (see Figs. 4, 5, and 8).

Fig. 4.

Ca²⁺ responses of CB5 axon terminals to stimulus movement direction. A, left: cross-sectional view of a reconstructed CB5 cell (arrow) and a rod BC (arrowhead) in a ChAT-Cre/tdTomato (red) mouse retina. Top right: tangential two-photon image of the CB5 cell on left, showing cofasciculation between CB5 axon terminal branches and ChAT processes (yellow). Bottom right: ROIs drawn from individual terminal boutons of the CB5 (2–5) and the distal axon stalk of the rod BC (1). Scale bars, 10 μm. B, left: examples of Ca²⁺ signals (ΔF/F) from 5 ROIs shown in A, in response to a light bar moving in 4 orthogonal directions (1: rod BC, 2–5: CB5). Black traces, average responses from 5 trials; gray traces, responses from individual trials. Right: mean amplitude and SD (error bars) of responses shown on left. No statistically significant difference was found between responses to opposite movement directions for any of the ROIs on this CB5 terminal and rod BC axon. C: summary of directionally selective index (DSI) values measured from rod BC axon stalks (9 cells), CB5 terminal boutons that contacted ChAT processes (61 boutons from 7 cells), and CB5 boutons that did not contact ChAT processes (35 boutons from 7 cells). No statistically significant difference was found among the 3 groups. D, top: 2-photon image of an ON-OFF DSGC (filled with sulforhodamine 101, red) and ON BC axon terminals (transfected with GCaMP3, green) in the wild-type mouse retina. Bottom: expanded view of the boxed region at top, showing a segment of ON-OFF DSGC dendrite being contacted by the terminal branches (left) of a CB5 cell (as revealed by 3D reconstruction, center) whose terminal boutons are specified by ROIs (right). E: Ca²⁺ responses from 5 example CB5 boutons in D (bottom right), showing the averaged (black traces) and individual responses (gray traces, 5 trials) to moving bar stimulation in 4 orthogonal directions, with mean response amplitudes and SDs (error bars) shown on right. No statistically significant difference was found between responses to opposite movement directions for all but 1 ROI (not shown) on this CB5 cell. F: DSI measured from rod BC axon stalks (9 cells), CB5 boutons that contacted ON-OFF DSGC dendrites (6 boutons from 2 cells), and all CB5 boutons at the focal plane of nearby ON-OFF DSGC dendrites (49 boutons from 2 cells). No statistically significant difference was found among the 3 groups. Scale bars, 10 μm.

Fig. 5.

Effects of GABA receptor antagonists on Ca²⁺ responses to moving bar stimulation. A: 2-photon image of a CB5 axon terminal showing 6 representative ROIs. B: Ca²⁺ responses from the 6 ROIs in A to a light bar moving in 4 orthogonal directions in control and 50 μM SR95531 + 100 μM TPMPA (black, averaged responses; gray, overlay of 5 individual responses). C: DSI in control and 50 μM SR95531 + 100 μM TPMPA, showing no statistically significant difference after blockade of GABA_A/C receptors (69 boutons in 5 CB5 cells, P = 0.13).

Fig. 8.

Receptive field properties of GCaMP3-expressing CB7 cells. A: 2-photon image of a CB7 axon terminal (green) and ChAT processes (red) in a wholemount view (top) of a ChAT-Cre/tdTomato mouse retina, with ROIs drawn over various terminal segments (middle). Bottom: cross-sectional view of the same axon terminal. Arrow indicates the terminal of the CB7 cell; arrowhead points to terminal processes of a nearby CB5 cell. Scale bars, 10 μm. B: Ca²⁺ responses from different terminal regions (ROIs 1–8 in A) to a light bar moving in 4 orthogonal directions (black, averaged response from 5 trials; gray, overlay of responses from individual trials). C: comparison of DSIs measured from rod BC axons and CB7 boutons (contacting and not contacting ChAT processes). No statistically significant difference was found among the 3 groups. D: cross-sectional view of another CB7 cell (terminal indicated by arrow; arrowhead indicates the terminal of an adjacent CB5 cell) (top) and wholemount view of its axon terminal with 3 randomly chosen ROIs (bottom). Scale bars, 5 μm. E: Ca²⁺ signals from the axon terminal boutons of the CB7 in D, in response to a flash of a center light spot of increasing diameter (gray, individual trials; black, average from 3 trials; yellow, light spots). F: normalized response amplitude vs. light spot diameter for CB7 cells. Thin color lines, averaged responses from all boutons in a cell (error bars, SD); thick black lines, averaged spatial profile of responses from 2 cells.

RESULTS

CB5 cells do not receive fast GABAergic synaptic feedback from SACs.

Because CB5 cells are thought to provide major excitatory inputs to DSGCs (Helmstaedter et al. 2013), we tested directly whether SACs make functional GABAergic feedback synapses onto CB5 cells by photoactivation of SACs in ChAT-Cre/ChR2-YFP mice while recording from CB5 cells with patch clamp in the wholemount retina. To focus on potential direct GABAergic inputs from SACs, we blocked other major synaptic receptors in the IPL and photoreceptor-driven light responses in the retina with a cocktail of pharmacological agents containing (2S)-2-amino-4-phosphonobutanoic acid (l-AP4, 20 μM), 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX, 80 μM), 3-(2-carboxypiperazin-4-yl)propyl-1-phosphonic acid (CPP, 20 μM), hexamethonium (HEX, 300 μM), and strychnine (1 μM), which are blockers of mGluR6, non-NMDA, NMDA, nicotinic, and glycine receptors, respectively. As a control, we first recorded from ON-OFF DSGCs to verify that GABAergic synaptic release from SACs could be evoked and detected under our experimental condition. Figure 1, A and B, show an example of a patch-clamped ON-OFF DSGC, with its dendrites (filled with Alexa Fluor 594; Fig. 1A) costratifying with the two bands of ChR2-expressing SAC processes (green, Fig. 1A, bottom). In 10 of the 10 ON-OFF DSGCs tested in the wholemount retina, photoactivation of ChR2 reliably evoked a large outward current in the DSGC at 0 mV (∼E_Cation), which could be completely blocked by the GABA_A receptor antagonist SR95531 (50 μM, n = 5) (Fig. 1B). However, in recordings from 11 CB5 cells (4 in wholemount retina, 7 in retinal slices; Fig. 1C), we did not detect any fast synaptic currents in response to photoactivation of ChR2-expressing SACs (Fig. 1D), suggesting a lack of functional GABAergic synaptic feedback from SACs to CB5 cells. In 2 of the 11 CB5 cells tested, a small (<10 pA) and slowly rising outward current was detected at 0 mV after photoactivation of SACs, which appeared to be a result of nonsynaptic GABA diffusion (data not shown; see discussion).

Fig. 1.

Lack of fast GABAergic synaptic feedback from starburst amacrine cells (SACs) to CB5 cells. A, top: wholemount view [maximal-intensity projection from a stack of 2-photon images taken around the ON choline acetyltransferase (ChAT) band] of an ON-OFF direction-selective ganglion cell (DSGC; filled with Alexa Fluor 594 during whole cell patch clamp, red) and the ChAT/ChR2 plexuses (green) in a ChAT-Cre/ChR2-YFP mouse retina. Bottom: cross-sectional view of the same ON-OFF DSGC showing dendritic overlap with the ChAT/ChR2 bands (yellow). B: responses of the ON-OFF DSGC to photo activation of SACs (timing of light flashes shown at top), recorded in the wholemount retina under whole cell voltage clamp at 0 mV, showing outward GABAergic synaptic currents that could be completely blocked by 50 μM SR95531. Control solution: ACSF supplemented with 20 μM (2S)-2-amino-4-phosphonobutanoic acid (l-AP4), 80 μM 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX), 20 μM 3-(2-carboxypiperazin-4-yl)propyl-1-phosphonic acid (CPP), 300 μM hexamethonium (HEX), and 1 μM strychnine. C, left: wholemount view 2-photon image of CB5 terminals (red) and ChAT/ChR2-YFP processes (green). Right: cross-sectional view of the CB5 cell (red) and ChAT/ChR2-YFP processes (green). D: voltage-clamp recording (at 0 mV) of the CB5 cell in C in wholemount retina, showing no detectable response to the activation of ChR2-expressing SACs under the same condition as in B. Scale bars, 10 μm.

Two-photon imaging of light-evoked Ca²⁺ responses from ON BCs.

To characterize light-evoked calcium responses from individual BCs in the wholemount mouse retina, we transfected the retina with a DNA plasmid encoding the calcium sensor GCaMP3 under the control of the mGluR6 promoter. Transfected cells were identified under two-photon imaging to be exclusively ON types of BCs, with dendrites ramifying in the outer plexiform layer (OPL) and axons terminating in the proximal half (sublamina b) of the IPL (Fig. 2A). Calcium imaging was made from areas where GCaMP3-expressing BCs could be distinguished individually and reconstructed in 3D from a z-series of two-photon images. Identification of ON BC types was made based mainly on the depth and pattern of their axon stratification in the IPL, with the fluorescent cholinergic bands in ChAT-Cre/tdTomato mice as a reference (Fig. 2A). Cells with morphologies resembling those of CB5, CB6, CB7, CB8, and rod BCs in the mouse retina (Ghosh et al. 2004; Pignatelli and Strettoi 2004; Wassle et al. 2009) were found in our samples of GCaMP3-expressing BCs, but rod BCs and CB5 cells were most frequently encountered. While rod BC axons passed through the cholinergic bands to terminate at the proximal border of the IPL (Fig. 2A), both CB5 and CB7 cells stratified near the ON ChAT band. CB5 cells stratified narrowly in sublamina 3 near the distal margin of the ON ChAT band (Fig. 2A), with some of their terminal processes descending further into the ChAT band and making close contacts with ON SAC processes (Fig. 2A and Fig. 3A). On the other hand, CB7 cells ramified in sublamina 4 just below the ON ChAT band, with some of their terminal processes overlapping with the ON SAC processes (Fig. 2A and Fig. 7). Figure 2B shows typical Ca²⁺ signals from axon terminals of rod BC, CB5, and CB7 cells in response to a flashed light spot (80 μm in diameter) presented in the receptive field center. At the offset of the light flash, a dip below the baseline fluorescence intensity was often observed from CB5 cells. The light-evoked fluorescence responses were most robust in the axon terminal arbor (ΔF/F up to 400%), weaker in the distal axon stalk (near the terminal arbor), and undetectable in the proximal axon stalk (near the soma), soma, and dendrites (Fig. 2, C–E).

Fig. 2.

Light-evoked responses of GCaMP3-expressing ON bipolar cell (BC) types under 2-photon imaging. A, left: cross-sectional view of reconstructed ON BCs (left) in a ChAT-Cre/tdTomato mouse retina, showing GCaMP3-expressing ON BCs (green) and ON and OFF cholinergic bands (red). Right: examples of a rod BC (left), a CB5 (center), and a putative CB7 (right). Scale bars, 10 μm. OPL, outer plexiform layer; INL, inner nuclear layer; IPL, inner plexiform layer; GCL, ganglion cell layer. B: light-evoked Ca²⁺ signals measured from the axon terminal of a rod BC (left), a CB5 (center), and a CB7 cell (right) in response to an 80-μm center light spot (duration of light stimulation indicated in yellow). Black, averaged responses from 5 trials; gray, responses from individual trials. C: cross-sectional view of 3 ON BCs reconstructed from a z-stack of 2-photon images taken from the IPL to the OPL, with a CB5 cell shown in the center. D: 2-photon images of the CB5 cell in C, taken at focal planes indicated by the dashed lines in C, showing (from top to bottom) the dendrites, soma, proximal axon stalk, distal axon stalk, and axon terminal of the cell in wholemount view. Scale bars in C and D, 5 μm. E: fluorescence responses to a flash of 80-μm center light spot recorded from the regions of interest (ROIs) drawn in the corresponding frames in D, showing detectable responses at the axon terminal and distal axon stalk but not at the proximal axon stalk, soma, or dendrites. Black, averaged responses from 10 trials; gray, superimposed responses from 10 individual trials.

Fig. 7.

CB7 cells provide direct excitatory synaptic input to, but do not receive direct GABAergic feedback from, SACs. A, left: dual patch clamp from a pair of CB7 and ON SACs (both filled with Alexa Fluor 594) in the wholemount ChAT-Cre/ChR2-YFP mouse, showing CB7 terminal boutons (in white box) and SAC dendrites on a background of ChAT/ChR2 plexuses (green) (image reconstructed from maximal-intensity projection from a stack of 2-photon images taken around the ON ChAT band). Center: blown-up view of the area inside the white box on left, showing CB7 terminal boutons and the dendrites of the patch-clamped SAC (arrows). Right: cross-sectional view of the CB7 (reconstructed from a z-stack of 2-photon images, red) and ChAT/ChR2-YFP bands (green). B: dual voltage-clamp recording in ACSF supplemented by 20 μM l-AP4, showing voltage-gated currents in the CB7 (middle) and excitatory synaptic currents in the SAC (bottom), in response to voltage steps from −70 mV to −20, −10, 0, +10, and +20 mV (top) applied to the SAC. C: cross-sectional view (reconstructed from z-series 2-photon images) of a patch-clamped CB7 cell (red) in the wholemount ChAT-Cre/ChR2-YFP (green) retina. D: photoactivation of SACs with light flashes (top) failed to evoke any synaptic current (bottom) in the CB7 cell at 0 mV in ACSF supplemented with 80 μM CNQX, 20 μM CPP, 20 μM l-AP4, and 300 μM HEX. Scale bars, 10 μm.

Ca²⁺ responses from individual branches of CB5 axon terminals.

Because CB5 cells are thought to provide major excitatory input to DSGCs (Helmstaedter et al. 2013), we first examined light-evoked and spontaneous Ca²⁺ activities in different regions of the CB5 axon terminal to determine whether the activities were coordinated over the entire axon arbor or compartmentalized as isolated events in local terminal segments. Figure 3 shows a cross-sectional view of several reconstructed ON BCs (including CB5, CB8, and rod BCs) in a ChAT-Cre/tdTomato mouse retina (Fig. 3A). In a tangential (wholemount) two-photon image (Fig. 3B), the axon terminal of the center CB5 cell (cell 1) could be clearly identified (green/yellow processes), showing cofasciculation/overlap with the SAC processes (red). In the same image, a small terminal branch of a neighboring CB5 cell (cell 2, in blue box) and the axon stalk of a putative CB8 (cell 3, in red box) were also identified, which were verified by tracing the processes back to their terminal arbors and somas in z-series two-photon images. In response to the onset of IR laser scanning, various ROIs in the two CB5 cell terminals (Fig. 3C) displayed prominent Ca²⁺ increases, which declined to a low steady-state level within ∼15–20 s under continuous laser scanning (Fig. 3D). Subsequent stimulation of the retina with a visible light spot at different locations of the receptive field (Fig. 3D, inset) evoked reliable Ca²⁺ responses from all ROIs of the two CB5 cells (Fig. 3E). While the light responses from different ROIs had different amplitudes (ΔF/F), the kinetics of the responses was remarkably similar among all ROIs of the same CB5 cell and between ROIs of the two neighboring CB5 cells. When normalized by the peak amplitude of their response to the onset of IR laser scanning, the visible light-evoked responses in all ROIs became nearly completely superimposable (Fig. 3E), suggesting that all regions of the CB5 axon arbor responded to a light stimulus not only with the same kinetics but also with the same relative amplitude. Notably, this close correlation among the responses of individual terminal regions remained the same when the light spot was presented at different positions to activate different parts of the receptive field center and surround. Similar results were also obtained in 10 other CB5 cells with larger (200- to 500-μm diameter) light spots (data not shown). While the two neighboring CB5 cells in Fig. 3 showed very similar response kinetics, the distal axon of the neighboring CB8 cell (cell 3, Fig. 3B), which passed through but did not stratify at the image plane, showed very different responses (Fig. 3E).

In addition to light-evoked responses, CB5 axon terminals occasionally showed spontaneous Ca²⁺ activity under continuous IR laser scanning and constant background illumination (e.g., around time points 57 s and 170 s in Fig. 3, D and E). The spontaneous Ca²⁺ signals varied in amplitude and waveform from event to event, but each event had a similar waveform and similar normalized amplitude in all ROIs, indicating that they were highly correlated among various regions of the axon terminal. Very rarely, we did observe sporadic spontaneous activities that appeared in one, but not the rest, of the ROIs of a CB5 axon arbor (Fig. 3, F and G). Even in such cases, light-evoked responses were still highly correlated among all ROIs, suggesting that most, if not all, activities observed under our recording conditions were not restricted to compartmentalized terminal regions (see discussion).

Responses to movement direction.

We next investigated whether responses of individual boutons on a CB5 axon terminal had any direction selectivity, either along a common preferred-null axis or along individual preferred-null axes. Since CB5 axon arbors stratify near, but not exclusively on, the ON cholinergic band, we examined all the boutons within each CB5 terminal arbor, including both terminal branches that contacted ON SAC processes and those that did not. We also injected fluorescent dyes in ON-OFF DSGCs and examined CB5 boutons that contacted ON-OFF DSGC dendrites. In all these cases, Ca²⁺ responses from CB5 boutons were found to have no clear directional preference. Figure 4 shows Ca²⁺ signals in CB5 axon boutons that contacted SAC (Fig. 4, A and B) and ON-OFF DSGC (Fig. 4, D and E) dendrites in response to a light bar moving in four orthogonal directions. Although the response amplitude varied somewhat among different movement directions, the mean responses to opposite movement directions were not statistically different for the majority of boutons tested (140 of 151 boutons, P > 0.05) (Fig. 4, B and E). A small number of boutons (11 of 151) showed small, but statistically significant (P < 0.05), differences in response amplitude to at least one pair of opposite moving directions. However, the differences were small and the DSI (see below) calculated from these boutons was statistically indifferent from that measured from the rest of the boutons (P = 0.29).

As shown in Fig. 4, C and F, the mean DSI was 0.053 ± 0.029 (mean ± SD) (n = 61 boutons from 7 cells) for those CB5 boutons that contacted the ON ChAT processes; 0.054 ± 0.026 (n = 35 boutons from 7 cells) for boutons that did not contact the ON ChAT processes; 0.049 ± 0.018 for boutons that contacted dye-injected DSGC dendrites (n = 6 boutons from 2 cells); and 0.061 ± 0.031 for boutons that did not contact dye-injected DSGC dendrites (n = 43 boutons from the same 2 cells). Because rod BC axons are not expected to be directionally selective and can serve as a reference for DSI, we simultaneously recorded the light responses of rod BC axons at the points where they passed through the same image plane of CB5 terminals and ChAT (or ON-OFF DSGC) processes (Fig. 4, A and D). The mean DSI from rod BC axons was 0.043 ± 0.032 (n = 9 cells), not significantly different from that of CB5 boutons (P = 0.57 for boutons contacting ON-OFF DSGC, P = 0.22 for all boutons) (Fig. 4, C and F). In contrast, the DSI of GCaMP3 signals recorded under the same condition from DSGC somas was significantly different: 0.79 ± 0.14 (n = 2, P < 0.05; data not shown).

Because the previously reported directional asymmetry in the glutamate current recorded from ON-OFF DSGCs under voltage clamp was eliminated by blocking GABA receptors (Fried et al. 2005, 2006; Lee et al. 2010; Park et al. 2014), we tested whether GABA receptor antagonists would reduce the DSI values measured from CB5 terminals and reveal a weak, GABA receptor-dependent directional selectivity. As shown in Fig. 5, blocking GABA_A and GABA_C receptors (with 50 μM SR95531 and 100 μM TPMPA) enhanced the response amplitude by 72 ± 20% (n = 69 boutons from 5 cells) but did not have a statistically significant effect on the DSI of CB5 responses to moving bar stimulations [the DSI actually increased from 0.061 ± 0.034 to 0.070 ± 0.042 (n = 69 boutons from 5 cells), but the difference was statistically insignificant, P = 0.13]. Thus there was not a resolvable GABA-dependent direction-selective component in CB5 responses.

Two types of receptive field structures for cells with CB5 morphology.

To determine whether there are functionally distinct subtypes of CB5 cells in the mouse retina, we investigated the receptive field properties of morphologically similar CB5 cells by measuring the light-evoked Ca²⁺ responses in their axon terminals. When the center-surround receptive structure of CB5 cells was tested with a light spot of increasing diameter, we found two distinct types of receptive field properties. In the first type (which we named CB5₁), the response amplitude increased to its peak value at a light spot diameter of ∼100 μm and then slowly declined as the spot diameter increased further from 100 μm to 1,400 μm, such that the response amplitude remained at 59% ± 13% (n = 8) of its peak value when the light spot diameter reached 500 μm and at 50.7% ± 5.4% (n = 3) when the spot diameter increased to 1,400 μm (Fig. 6, A, C, E, and G). In the second type (named CB5₂), the response amplitude also reached a peak value at ∼100-μm spot diameter, but it then declined quickly as the light spot diameter further increased, such that the amplitude was reduced to 13.4 ± 6% (n = 6) of its peak value at a spot diameter of 300 μm and became nearly undetectable at a spot diameter of 500 μm (Fig. 6, B, D, F, and H), suggesting a much stronger inhibitory surround compared with the first type. Thus morphologically similar CB5 cells consisted of at least two functional subtypes, distinguishable on the basis of their center-surround receptive structure (see discussion).

Fig. 6.

Two subtypes of center-surround receptive field structures among CB5 cells. A and B, left: cross-sectional view of 2 CB5 cells reconstructed from z-series 2-photon images in ChAT-Cre/tdTomato mouse retinas. Right: tangential images of the 2 CB5 terminals at the focal plane of ON ChAT processes (red channel not shown), with ROIs randomly drawn over terminal boutons. Scale bars, 10 μm. C and D: Ca²⁺ signals from axon terminal boutons of the two CB5 in A and B, in response to a flash of a center light spot of increasing diameter (gray, individual trials; red, average from 5 trials; yellow, timing of light flash). E and F: response amplitude (normalized by the maximum value) vs. light spot diameter measured at each ROI in C and D (error bars: SD over 5 trials), showing weak (E) and strong (F) surround inhibitions, suggestive of 2 distinct types of receptive field structures. G and H: normalized response amplitude vs. light spot diameter for all CB5 cells with a weak (G) and a strong (H) inhibitory surround. Thin color lines, averaged responses from all boutons in a cell (error bars, SD); thick black lines, averaged spatial profile of responses from all cells in each subtype.

Notably, both of the above two functional subtypes of CB5 cells were tested for direction-selective responses to moving bar stimulations. For CB5₁ cells, the DSI was 0.055 ± 0.029 (n = 66 boutons, 4 cells). For CB5₂ cells, the responses to moving bar stimulation were considerably smaller in all directions, perhaps because of the stronger surround inhibition. The DSI for CB5₂ cells was 0.086 ± 0.038 (n = 65 boutons, 5 cells), slightly higher than that for CB5₁ cells, possibly in part because of a relatively larger variability associated with measuring smaller responses from CB5₂. Although the difference in DSI between the two subtypes was statistically significant (P < 0.0001), both subtypes had very low DSI values. In addition, application of SR95531 (50 μM) and TPMPA (100 μM), which is expected to eliminate directional asymmetry (Fried et al. 2002, 2005; Park et al. 2014), did not significantly reduce the DSI of CB5₂ cells (0.079 ± 0.036 in control, 0.069 ± 0.045 in SR+TPMPA; n = 15 boutons, 2 cells, P = 0.35). Thus we concluded that neither subtype was direction selective.

CB7 bipolar cells provide functional input to ON SACs but do not receive direct synaptic feedback from SACs or have directionally selective Ca² responses at axon terminals.

In addition to CB5 cells, CB7 cells also send many axon terminal processes to the ON ChAT/DSGC sublamina of the IPL. CB7 cells have been suggested to provide inputs to ON SACs and possibly ON-OFF DSGCs (Helmstaedter et al. 2013; Shi et al. 2011), but there has been no functional information about such inputs. We investigated whether CB7 cells provided functional excitatory input to the starburst network, whether they received GABAergic feedback from the starburst network, and whether their axon terminals generated directionally selective light responses. Paired patch-clamp recordings were made from CB7 cells and SACs in the wholemount mouse retina with pipettes containing Alexa Fluor 594 (Fig. 7A). The morphological identities of CB7 and ON SACs were confirmed at the end of the recording by two-photon imaging, which showed CB7 axon terminals ramifying near the proximal margin of the ON ChAT band and making contacts with ON SAC dendrites (Fig. 7, A and C). Depolarizing the CB7 cell with depolarizing voltage pulses (from −70 mV to −20 mV and above) evoked large synaptic responses in the postsynaptic SAC at −70 mV (∼E_Cl) (Fig. 7B), demonstrating the presence of functional excitatory synapses from CB7 to ON SACs.

To determine whether CB7 cells receive direct functional GABAergic feedback from SACs, we photoactivated SACs in the wholemount retina of ChAT-Cre/ChR2-YFP mouse, while recording from CB7 cells under voltage clamp in the presence of l-AP4 (20 μM), CNQX (80 μM), CPP (20 μM), HEX (300 μM) and strychnine (1 μM). In all seven CB7 cells recorded at 0 mV, we did not detect any currents in response to the activation of SACs (Fig. 7C), suggesting a lack of functional GABAergic feedback from SACs to CB7 cells.

To determine whether CB7 axon terminals generate direction-selective responses to moving light stimulation, GCaMP3-expressing CB7 cells in a ChAT-Cre/tdTomato mouse retina were studied under two-photon Ca²⁺ imaging in the same way as for CB5 cells (Fig. 8A). Ca²⁺ responses to a light bar moving in four orthogonal directions were measured from individual CB7 axon terminal regions (Fig. 8B). Average DSI measured from 23 boutons that contacted ON ChAT processes (n = 3 cells) was 0.07 ± 0.041 (Fig. 8C). Similar measurements from 22 boutons that did not contact ChAT processes yielded a similarly small average DSI of 0.062 ± 0.048 (Fig. 8C). These DSI values were not statistically different from that of rod BC axons (Fig. 8C, P = 0.28), indicating a lack of direction selectivity at CB7 axon terminals. The center-surround receptive field structure of CB7 axon terminals was also studied with flashes of a stationary light spot of increasing diameter (Fig. 8, D–F), revealing an inhibitory surround with a strength that was between those of CB5₁ and CB5₂ (Fig. 6).

DISCUSSION

The present study investigated axonal signal processing in BCs, which has been proposed to generate some degree of direction selectivity in BC axon terminals that synapse onto DSGCs. To explain the directionally asymmetric glutamate currents recorded from DSGCs, it has been hypothesized that axon terminals of certain cone BC types receive spatially asymmetric GABAergic input from SACs and generate direction-selective responses. Because there do not appear to be enough BC subtypes to subserve all types of ON-OFF and ON DSGCs, it has been further postulated that different regions of a BC axon terminal may make compartmentalized computations and synapse selectively onto multiple types of DSGCs, similar to the manner in which a SAC provides different directional inputs to multiple DSGCs (Euler et al. 2002; Fried et al. 2002, 2005; Hausselt et al. 2007; Lee and Zhou 2006; Vaney et al. 2012; Wei et al. 2011). We directly tested these hypotheses on axon terminals of mouse CB5 and CB7 cells. The results showed that 1) CB5 cells do not receive fast GABAergic synaptic feedback from SACs; 2) Ca²⁺ responses are well correlated in kinetics and amplitude among different regions of a CB5 axon terminal; 3) CB5 axon terminals respond to moving light stimulation with no detectable direction selectivity; 4) CB5 cells consist of two different functional subtypes with distinct receptive surround structures, with neither subtype showing directional selectivity; 5) CB7 cells make functional glutamatergic synapses onto displaced SACs but receive no direct GABAergic synaptic feedback from SACs; and 6) CB7 axon terminals also do not have detectable direction selectivity.

Synaptic connectivity between SAC and CB5 and CB7 cells.

Our study is the first to functionally test whether SACs make direct GABAergic feedback synapses onto BC terminals. We found no evidence for a direct and fast GABAergic synaptic input from SACs to CB5 cells. There has been no report from EM studies of feedback synapses from SACs to BCs in the mouse retina thus far. EM studies of the rabbit retina also do not find any synapses from SACs to BC terminals (Famiglietti 1991), but such synapses have been reported in the primate retina (Yamada et al. 2003). In 2 of the 11 CB5 cells recorded, a small, slowly rising TPMPA-sensitive outward current was detected upon activation of ChR2-expressing SACs (data not shown). This small current was very different from the synaptic GABA currents recorded from DSGCs under the same stimulation condition, and might represent small, nonsynaptic GABA diffusion from SACs to a small subset of CB5 cells. Because BCs were not transfected with GCaMP3 or tested for light responses in these experiments, we do not know the receptive field properties of the CB5 cells that showed this small current. Overall, our results suggest that CB5 cells do not receive synaptic GABAergic feedback from SACs, though we cannot completely rule out the possibility that some sparse inputs to CB5 axon terminals might be too small to be detected by patch clamp at the soma.

Our double-patch recordings present the first physiological evidence that CB7 cells make functional glutamatergic synapses onto ON SACs. Because of the technical challenges in making such paired recordings, the detailed physiology of this synapse remains to be investigated in the future. We showed that none of the CB7 cells tested had any direct response to photoactivation of ChR2-expressing SACs. The lack of direct GABAergic synaptic feedback inhibition from SACs to CB5 and CB7 cells provides a strong argument against the hypothesis that SACs provide spatially asymmetric GABAergic inhibition onto presynaptic BCs to produce direction-selective release of glutamate.

Electrotonic structure of CB5 terminals.

BC axon terminals are generally thought to be electronically compact because of their limited physical size. However, a recent study found that some BCs in the salamander retina can send distinct signals to different target ganglion cells, raising the possibility of compartmentalized computation in BC axon terminals (Asari and Meister 2012). We now tested this possibility in mouse CB5 cells by comparing the kinetics and relative amplitude of Ca²⁺ signals from various regions of the axonal arbor. Our results showed that light-evoked Ca²⁺ signals were well correlated among different parts of the CB5 axon terminal, suggesting that the axon terminal behaves largely as a compact signal processing unit, consistent with the spatially compact geometry of CB5 axon terminals (Ghosh et al. 2004). We found that even the spontaneous Ca²⁺ activities in CB5 terminals were nearly always correlated under our recording condition, although we did observe isolated small local spontaneous events on rare occasions (Fig. 3G). It should be pointed out that, overall, spontaneous activities were seldom observed in our recordings, and we could not rule out the possibility that this was due, in part, to the illumination condition (constant IR laser scanning and the presence of a weak background light) and/or the limitations in GCaMP3 sensitivity and kinetics.

Receptive field properties of CB5 and CB7 cells.

Our finding that the Ca²⁺ responses from CB5 axon terminals are not directionally selective to moving bar stimulation is consistent with a recent report (Yonehara et al. 2013). Because Yonehara et al. (2013) only studied CB5 terminals that were transsynaptically coupled to ON DSGCs, which are different from ON-OFF DSGCs in response kinetics, speed tuning, and possibly BC inputs (Ariel and Daw 1982; Sivyer et al. 2010), it is necessary to determine whether CB7 terminals, which are reported to provide input to ON-OFF DSGCs (Shi et al. 2011; but also see Helmstaedter et al. 2013), are directionally selective. It is also necessary to rule out the possibility that subpopulations of CB5 cells that synapse specifically onto ON-OFF (but not ON) DSGCs are directionally selective. Our results demonstrate for the first time that CB7 cell terminals also lack directional Ca²⁺ responses. Moreover, we discovered that cells with a similar CB5 morphology under light microscopy actually consisted of two different types of receptive fields (see below), and that neither type showed direction selectivity in Ca²⁺ responses. It should be pointed out that, because glutamate release from BC terminals may have a nonlinear (cooperative) dependence on Ca²⁺ concentration, there is a possibility that a small directional difference in Ca²⁺ response, which could be masked by the noise and/or the response variability inherent to the imaging experiment, might be sufficient to produce a directional difference in glutamate release. While this possibility cannot be presently excluded, the directionally asymmetric Ca²⁺ response observed in SAC dendrites argues against it (Euler et al. 2002; Lee and Zhou 2006; Yonehara et al. 2013). Because the DSI values calculated for Ca²⁺ responses in BC axon terminals are expected to be very different from those calculated from DSGC spike responses, we used the DSIs measured from the Ca²⁺ responses of rod BC axons and ON-OFF DSGC somas as references and also compared the DSI values from the same CB5 terminals before and after blocking GABA receptors. We conclude that the low DSI values at CB5 and CB7 axon terminals most likely represent a genuine lack of direction selectivity.

Taken together, our study tested all three assumptions of the BC direction selectivity model. In light of our findings of 1) the electrotonically compact axon terminal structure, 2) the lack of direct GABAergic synaptic feedback from SACs, and 3) the lack of direction-selective Ca²⁺ responses from all three functional types of BC terminals that ramify in the direction-selective strata of the IPL (CB5₁, CB5₂, and CB7), we conclude that the BC axon terminals synapsing onto DSGCs and SACs are not directionally selective. This conclusion is consistent with recent measurements of glutamate release using the glutamate sensor iGluSnFR (Marvin et al. 2013), although these measurements were not made specifically from identified CB5 or CB7 synaptic boutons (Park et al. 2014; Yonehara et al. 2013). Thus the directionally asymmetric glutamate currents measured from DSGCs under whole cell voltage clamp are not attributable to directionally selective release of glutamate from BCs but may be caused by other factors, such as space-clamp inadequacy in whole cell patch-clamp recordings of DSGCs (Poleg-Polsky and Diamond 2011).

Functional subtypes of CB5 cells.

Our finding of two kinds of receptive field structures of CB5 cells provides a new functional subclassification of BC types based on receptive field properties. Because CB5 cells have a nearly twofold higher density and cone-to-BC ratio than most other cone BC types in the mouse retina, two subpopulations of CB5 cells, CB5a and CB5b, have been proposed (Wassle et al. 2009). However, it has been unclear whether functionally different CB5 subtypes actually exist in the mouse retina. We now show that CB5 cells can be functionally subdivided into two subtypes, CB5₁ and CB5₂, on the basis of their different receptive field surround properties. Because CB5a and CB5b cells are not yet specifically defined in the mouse retina (Wassle et al. 2009), a possible correspondence between CB5₁/CB5₂ and CB5a/CB5b remains to be established. In preliminary experiments, we found that blocking GABA_A/C and glycine receptors could remove the strong surround inhibition in CB5₂ cells, such that the response amplitude remained similar as the light spot diameter increased from 100 to 500 μm (data not shown). Thus it seems likely that the differences in the strength and spatial extent of the surround inhibition between CB5₁ and CB5₂ are largely due to different synaptic connections with amacrine cell types. A recent EM study also found that CB5 cells could be divided into three subtypes (CB5a, CB5b, and XBC) based on their connectivity with ganglion cells (Helmstaedter et al. 2013). Together these results indicate that BCs terminating at a similar IPL depth may form different functional circuits and have different receptive fields. For example, ON and ON-OFF DSGCs ramify at a similar depth of sublamina b (Ackert et al. 2009; Dong et al. 2004; Sun et al. 2006) but receive kinetically different glutamate inputs and have very different light response kinetics and speed tuning properties (Ackert et al. 2009; Ariel and Daw 1982; Sivyer et al. 2010; Weng et al. 2005). Future studies will determine whether CB5₁ and CB5₂ preferentially target one over the other type of DSGC. Our finding of functionally different subtypes of CB5 cells suggests the possibility that other currently identified BC types (especially CB3) may also comprise multiple functional subtypes with distinctive receptive field structures, and that these additional BC subtypes may increase the number of parallel visual channels in the retina and the capacity of synaptic computation in the IPL.

GRANTS

This work was supported by National Eye Institute Grants R01 EY-17353 (Z. J. Zhou) and R01 EY-10894 (Z. J. Zhou) and an unrestricted grant from Research to Prevent Blindness Inc. to the Department of Ophthalmology and Visual Science, Yale University.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: M.C., S.L., and Z.J.Z. conception and design of research; M.C., S.L., and S.J.P. performed experiments; M.C., S.L., and Z.J.Z. analyzed data; M.C., S.L., and Z.J.Z. interpreted results of experiments; M.C., S.L., and Z.J.Z. prepared figures; M.C., S.L., and Z.J.Z. drafted manuscript; M.C., S.L., S.J.P., L.L.L., and Z.J.Z. edited and revised manuscript; M.C., S.L., S.J.P., L.L.L., and Z.J.Z. approved final version of manuscript.