Direction-Selective Circuitry in Rat Retina Develops Independently of GABAergic, Cholinergic and Action Potential Activity

The ON-OFF direction selective ganglion cells (DSGCs) in the mammalian retina code image motion by responding much more strongly to movement in one direction. They do so by receiving inhibitory inputs selectively from a particular sector of processes of the overlapping starburst amacrine cells, a type of retinal interneuron. The mechanisms of establishment and regulation of this selective connection are unknown. Here, we report that in the rat retina, the morphology, physiology of the ON-OFF DSGCs and the circuitry for coding motion directions develop normally with pharmacological blockade of GABAergic, cholinergic activity and/or action potentials for over two weeks from birth. With recent results demonstrating light independent formation of the retinal DS circuitry, our results strongly suggest the formation of the circuitry, i.e., the connections between the second and third order neurons in the visual system, can be genetically programmed, although emergence of direction selectivity in the visual cortex appears to require visual experience.


Introduction
Image motion is an important survival cue to animals. Among fewer than 20 types of retinal ganglion cells (RGCs) in the mouse [1], three are devoted to coding motion directions [2,3,4]. The ON-OFF direction selective ganglion cells (DSGCs) respond strongest when images move in a particular preferred direction, but very weakly when they move in the opposite null direction [5]. The mechanism for direction selectivity (DS) is a spatially asymmetrical inhibition [6] that DSGCs only receive inhibitory inputs from starburst amacrine cells (SACs) spatially offsetting to the direction in which image motion induces very little or no response, providing a physical entity for the asymmetrical inhibition [7]. The output of SACs is direction selective (DS) [8], produced cell-autonomously [9] or by reciprocal inhibition between overlapping SAC processes [10]. It is not known how the selective connections are formed between dendrites of DSGCs and the processes of SACs extending centrifugally in coincidental null directions. Recent findings have ruled out involvement of light driven activities and bipolar cell outputs [11,12,13].
SAC contains both acetylcholine (ACh) and GABA [14,15], and GABA plays a critical role in the computation of motion direction: blocking GABA receptors abolished DS [16] and the asymmetrical responses in its varicosities are mediated by reciprocal GABA inhibition [10], in addition to cell autonomous properties [9]. The enzyme synthesizing GABA is present very early in development [17], when the synchronous activities are modulated by GABA and glycine [18]. Activating GABA receptor induces an inward current [19,20] that becomes outward around postnatal day 6 (P6) with the expression of the potassium-chloride cotransporter, KCC2 [21,22]. It has been shown recently that direction selective circuitry develops normally after blocking GABA transmission in the 2 nd week of development [23]. It is not clear if activation of GABA receptors is necessary for the initial formation of contacts between SACs and DSGCs.
Spontaneous, synchronized propagating activities in the early developing retina [24], simultaneously activating RGCs and their target neurons, have been suggested to play a role in the formation of eye specific lamination in the lateral geniculate nucleus [25,26,27], and the precise geniculocortical mapping [28,29,30], although controversial evidence existed [31]. During this period, SACs express voltage gated sodium channels [32] and are activated synchronously with RGCs by propagating activities [33], it is conceivable that the wave-induced synchronous activities provide a possibility for the DSGCs and SACs to form selective connections through a spike timing dependent plasticity (STDP) [34], a Hebbian mechanism. From postnatal day 0 (P0) to P11, ACh release [35,36] triggered by pace-making activity of SACs [37] is critical for propagating activities. Normal DS circuitry has been observed in the nicotinic ACh receptor b2 subunit deficient mice in which retinal synchronous activities were diminished [13], arguing that synchronized activities are not important. A recent paper showed increased frequency of synchronous activity in the same mouse line [38] and therefore, strengthened STDP.
In this study, we used intraocular injection of pharmacological reagents to block the GABAergic activity, ACh driven synchronous activity or/and Na + dependent spiking activity, and demonstrated the morphology, physiology and the DS circuitry develop independently of GABAergic, cholinergic or spiking activity. Our findings strongly suggest a genetic mechanism in this circuitry formation.

Results
DSGCs are observed in the rat retina as early as P13, showing very similar dendritic morphology and physiology to their counterparts documented in the rabbit and mouse (Fig. S1), confirming that DS was present in the rat retina at the onset of the light sensitivity, similar to what was reported in the mouse [12].

Blocking GABA A receptors does not affect the development of DS circuitry
We first tested if GABA activities play any role in the formation of retinal DS circuitry by daily intravitreous injection of a potent GABA A receptor blocker, bicuculline methiodide (BMI) [39], from birth to the day before experiment (P15-18). We verified the effectiveness of daily injection from three aspects. First, retinal geniculate projection was severely disrupted, showing expanded contra-and ipsi-projection of the injected eye ( Fig. 1A-E), presumably due to increased activities induced by GABA blockade. Second, vitreous extracts from one eye injected 24 h earlier, completely blocked inhibitory inputs to a DSGC in another retina despite the dilution in the perfusion chamber (Fig. 1F), indicating the BMI effect lasted at least for 24 h, sufficiently long lasting for our daily injections. Third, low dose BMI completely blocked inhibitory input in a retina receiving 19 consecutive injections from birth (Fig. 1G), demonstrating no tolerance to repeated injections. DSGCs with normal light responses, direction selectivity and dendritic morphology could be found in retinas receiving 16 to 19 consecutive injections. Identified DSGCs responded to the onset and termination of a stationary flashing spot ( Fig. 2A). To a moving rectangle, clear DS responses could be observed ( Fig. 2B) with asymmetrical excitatory and inhibitory synaptic inputs (Fig. 2C). Bistratified dendritic morphology is illustrated in Fig. 2D, a section enlarged in 2E and F showing clear ramification in the ON and OFF sublaminae. Dendrites tightly costratified with cholinergic bands (Fig. 2G) in the inner plexiform layer (IPL). All together, fifteen DSGCs were recorded from ten BMI treated retinas. Data were summarized in Fig. 3. These experiments demonstrated that retinal DS circuitry develop normally without GABA mediated activities.
All four subtypes of DSGCs [40,41] were detected in BMI treated retinas (Fig. 4A). They were quite directional to drifting gratings with a contrast as low as 20% (Fig. 4B, C), although DSGCs in the treated retinas generally yielded weaker responses (Fig. 4B). The center-surround interaction was also intact as shown in histograms ( Fig. 4D) and in normalized arearesponse curve (Fig. 4E), responses increased with the size of spot initially and decreased when the spot entered the inhibitory surround.

Retinal DS circuitry develops independently of ACh induced synchronized activity
We then tested the roles of synchronized, ACh driven activity in the formation of retinal DS circuitry. We performed 14 to 19 consecutive daily intravitreous injections of a potent cholinergic agonist, epibatidine (Epb) which completely blocked synchronized activities with bath application [25,26,27,29]. Similar control experiments confirmed the sustaining effectiveness of Epb: 1) the retino-geniculate projection showed severe disruption with the ipsilateral projection of the treated eye to the LGN almost disappeared, and the ipsilateral axon terminals of the uninjected eye occupied a larger and diffused region ( Fig.  S2C-E). 2) Twenty four hours after an injection, the vitreous extracts from an injected eye completely blocked synchronized firing of a pair of neighboring RGCs of another early postnatal retina (P7), suggesting the intraocular concentration was sufficient to block synchronized activities throughout the treatment (Fig. S2F). 3) Bath application of 40 nM Epb blocked synchronous bursting of a pair of neighboring RGCs in a P9 retina receiving 9 consecutive intraocular injections, indicating no tolerance to Epb was caused by the extended treatment ( Fig.  S2G).
Retinal DS was still intact when patch clamp recordings were made from retinas receiving daily Epb injection. DSGCs exhibited responses to the onset and termination of a flashing spot (Fig. 5A), clearly directional responses to a moving rectangle (Fig. 5B), and asymmetrical synaptic inputs from voltage-clamp recordings (Fig. 5C). The bistratified dendritic arbor ( Fig. 5D-F) tightly costratified with cholinergic bands (Fig. 5G) in the IPL. Sixteen DSGCs were recorded from seven Epb treated retinas, the population data are shown in Fig. 3. These experiments showed that synchronized activities mediated by ACh are not essential for the development of retinal DS circuitry.

Spiking activity is not required for the development of DS circuitry
In Epb treated retina, the activity of about half of the RGCs was completely silenced, but for the remainders, the spontaneous activity was significantly increased [26,29,42]. Because spiking activity has been shown to be sufficient to induce retino-geniculate lamination [31], we further examined if spiking activity is sufficient to induce formation of the DS circuitry. We performed daily intravitreous injection of a voltage gated sodium channel blocker, tetrodotoxin (TTX) [30,43]. The sustained effectiveness of injected TTX was verified using same experiments as for BMI and Epb (Fig. S2E, H, I).
Almost normal DS responses can still be recorded from retinas receiving daily TTX injections. Fifteen DSGCs were recorded from eleven TTX treated retinas (Fig. 3) and detailed physiological and morphological data are shown in Figure 6 (A-G). These experiments demonstrated that spiking activity in both SACs and RGCs are not critical for the formation of retinal DS circuitry. Calcium imaging experiments showed that the ACh induced synchronized activities still existed in the TTX treated early postnatal retinas [44], in forms of synchronized elevation in intracellular calcium among neighboring cells. To completely rule out the involvement of any possible synchronized synaptic activity and spiking activity, we performed intraocular injection of TTX and Epb simultaneously. Normal DS responses were observed in 15 cells from 13 retinas (Fig. 3). Detailed physiological and morphological data are shown in Figure 7 (A-G). These experiments showed neither ACh driven synaptic activities nor synchronous or spontaneous spiking activities are required for the development of retinal DS circuitry.
The homeostatic properties of DSGCs are not changed in retinas treated by Epb and TTX (Fig. 4A-E), very similar to the results observed in BMI experiments.

Discussion
The present study is somewhat surprising because experience or activity dependent plasticity is involved in formation of neural connections in many sites of the visual system, such as the morphological modification of the neurons in the rat retina [45,46], the formation of eye specific lamination in the lateral geniculate nucleus and precise mapping of geniculocortical connections [25,26,27,28,29,30]. Experience is also critical in development of direction selectivity in visual cortical neurons [47].
On the other hand, it is not surprising because at the earliest stage of another sensory system, the olfactory system [48], the wiring of axon terminals to specific glomeruli is controlled by the type of olfactory receptor expressed in the olfactory neurons [49,50,51], although there is increasing evidence showing activity also plays a role in regulating the axon terminal targeting [52,53,54].
Most recent evidence showing two subtypes of ON-OFF DSGCs can be genetically labeled very early in development  [55,56] suggested these cells are destined to code a particular motion direction at the time or very soon after they are differentiated. However, the specification of cell type does not necessarily mean the completion of neural circuitry. In fact, several pieces of evidence suggest the formation of directional circuitry takes place in the first two postnatal weeks. When the morphology of RGCs were surveyed early in development, very few cells exhibited adult-like dendritic morphology [57,58], also the relationship between SACs and bistratified RGCs became increasing closer after P3 [59]. Most recently, direct evidence showing asymmetrical inhibition from SACs to DSGCs emerges after P7 [23].
Two recent studies reported that the postsynaptic GABAergic current in DSGCs induced by activating SACs is still symmetric at P7, but becomes asymmetric at P14 and older mice retinas [23]. Using optogenetic approach, Yonehara and colleagues demonstrated the asymmetrical inhibition between SACs and DSGCs is established around P8 [60]. With dual patch clamp recordings, the synaptic strength between null-side SACs and DSGC has been observed to significantly increase compared with the prefer-side during the second postnatal week [23], indicating a selective strengthening of inhibitory synapses between SACs and DSGCs. These findings are consistent with the previous report that DS responses emerge at the onset of light sensitivity at P11 [12].
Bipolar cell terminals reach the inner plexiform layer at P6, and the expression of vesicular glutamate transporter 1 (VGluT1) can be detected at P8 [61] coinciding with the establishment of the asymmetrical inhibition from SACs to DSGCs [60]. To address the role of spontaneous retinal activity during the establishment of DS circuitry further, Wei et al. injected muscimol or gabazine to activate or block GABAergic activity and showed that no change was observed in the asymmetric connection between SACs and DSGCs [23]. Our results ruled out the involvement of not only GABAergic, but also the cholinergic and Na + dependent action potential activities in the development of retinal DS circuitry. Furthermore, our results showed the GABA activity was not necessary for the initiation of contacts between SACs and DSGCs. Combined with previous findings ruling out light driven activities [11,12,13], the present study strongly suggests a genetic mechanism in the formation of selective connections between SACs and DSGCs.

Ethical approval
Use and handling of animals were strictly in accordance to the institutional guidelines and the Society for Neuroscience's policies on the use of animals and human subjects in neuroscience research [Approval Number: SYXK (SPF) 2007-119]. Animals were naturally mated and pups grew up with their mothers until the experimental day. They were maintained on a 12 hours dark/light cycle and the mothers were housed with libitum food (rat chow) and water (tap water). All pups were anesthetized and sacrificed under the guidelines mentioned above. And all efforts were made to minimize suffering.

Animals
Adult and new born SD rats were obtained from Charles River Laboratories and used in the experiments. Data were obtained from fifty nine early postnatal rats.

Intraocular Injections
We adopted procedures previously described for intraocular injection [25,30]. Before P8, animals were anesthetized with hypothermia and after P8 with ether. BMI (10 mg/mL, Sigma, St. Louis, MO, USA) in saline, or 0.2 mM TTX (Sigma, St. Louis, MO, USA) in 3.5 mM citrate buffer and/or 1 mM Epibatidine (Biomol, Plymouth Meeting, PA, USA) in 0.01 M PBS [26,39,43] made to the final volume of 1 mL, was injected daily intravitreally from P0 till the day before experiment (P13 to P18). The procedure was carried out under a dissecting microscope to insure clear visibility of the needle tip and every attempt was made to repeatedly inject through the same puncture in the sclera.
In the drug verification experiment, the eyeball received intraocular injection 24 h ago was enucleated and put in a small chamber. We collected as much vitreous humor as possible with a pipette after removing the cornea and lens, and applied the vitreous extracts onto the retina in an oxygenated recording chamber containing about 1 mL Ames medium.

Visualization of retino-geniculate projection
Cholera toxin B-Alexa Fluor 488 (Molecular Probes, Eugene, OR, USA, 2 mL of 0.5 mg/mL CTB in 0.01 M PBS) was injected into the treated eye and cholera toxin B-Alexa Fluor 594 ((Molecular Probes, Eugene, OR, USA) was injected into the intact eye [26]. Animals were sacrificed 3 days later and intracardially perfused with 4% paraformaldehyde (PFA) in 0.1 M PB. The brains were postfixed overnight in 4% PFA, cryoprotected in 30% sucrose solution (0.01 M PBS) and embedded in OCT. Coronal sections were cut at 50 mm on a cryostat (Leica CM1950). Twenty four sections were usually obtained throughout the LGN, and illustrations were selected from 9 to 16 th section. For each selected section, 10 images were captured in 30 mm depth. Areas of both the dLGN and the ipsilateral projection (both the treated and intact sides) were calculated for every single image. Images were captured using a confocal microscope (Olympus FluoView TM FV500). Brightness and contrast were adjusted using Image J 1.42 (National Institutes of Health) and Adobe Photoshop CS2 (Adobe Systensm Inc., San Jose, CA, USA).

Patch clamp recordings and data analysis
The procedures of patch clamp recording and data analysis have been described [4]. Briefly, rats were dark adapted for at least 2 hours then sacrificed with overdosed Sumianxin (mixture of xylazole, haloperidol and dihydroetorphine, 5 mL/kg), obtained from Institute of Veterinary Sciences, Academy of Military Medical Sciences, (Changchun, China). Eyes were enucleated immediately and a small cut was made at the temporal position (in order to orient the retina). Retinas were isolated and attached on black Millipore filters (AABP02500, Millipore, Billerica, MA, USA) with ganglion cell side up. Then the retinas were transferred into a recording chamber, continuously perfused with oxygenated bicarbonate-buffered Ames medium at 35uC at a rate of 5 mL/min.
After removing the inner limiting membrane with a pipette (filled with Ames solution), somas were exposed and RGCs had a medium sized elliptical soma were targeted, loose patch recording were carried out to identify ON-OFF DSGCs.
The closest stimulating direction to the vector sum of total spiking responses in 12 directions has been defined as preferred direction, and the direction selective index (DSI) is calculated as where Rp refers to responses in preferred direction and Rn indicates the responses in the opposite direction. While the tuning curves of responses to motion stimuli (red circles in Fig 2B, 5-7B and S1B) are fit with the von Mises distribution: Where R(x) is the response to the stimulus moving in a given direction x in radians, R max is the maximum response, m is the preferred direction in radians, and k is the concentration parameter used to calculate tuning width. The tuning width is estimated as the full width at half height of the von Mises fit:

Visual stimulus
Light stimuli about 0.35610 11 photons cm 22 s 21 was displayed on a monitor (Sony E230) and focused on the retina through microscope condenser. Four types of light stimuli were generated: (1) a 150 mm (in diameter) spot flashing for 1 s was used to determine the response polarity and receptive field, (2) a 1006500 mm rectangular moving parallel to its long axis at 19 deg s 21 over 1500 mm in 12 directions at 30u intervals was used to identify the directional preference, (3) flashing light spot of various diameter (from 50 to 800 mm, lasting for 1 s, repeating 3 times for each diameter) was used to check the center-surround interaction, and (4) moving square wave gratings (bright and dark, both are 200 mm width, 1 Hz) with different contrasts (from 0% to 100%) were used to detect the homeostasis of DSGCs. Percent contrast was defined as (L max 2L min )/(L max +L min ), where L max and L min were the maximum and minimum of luminance in this stimuli pattern.