Neuron-glia signaling in developing retina mediated by neurotransmitter spillover

Neuron-glia interactions play a critical role in the maturation of neural circuits; however, little is known about the pathways that mediate their communication in the developing CNS. We investigated neuron-glia signaling in the developing retina, where we demonstrate that retinal waves reliably induce calcium transients in Müller glial cells (MCs). During cholinergic waves, MC calcium transients were blocked by muscarinic acetylcholine receptor antagonists, whereas during glutamatergic waves, MC calcium transients were inhibited by ionotropic glutamate receptor antagonists, indicating that the responsiveness of MCs changes to match the neurotransmitter used to support retinal waves. Using an optical glutamate sensor we show that the decline in MC calcium transients is caused by a reduction in the amount of glutamate reaching MCs. Together, these studies indicate that neurons and MCs exhibit correlated activity during a critical period of retinal maturation that is enabled by neurotransmitter spillover from retinal synapses. DOI: http://dx.doi.org/10.7554/eLife.09590.001


Introduction
There is rich research history demonstrating the important role glial cells play in developing and maintaining neural circuits (Garrido, 2013). In the mature brain, glial cells provide structural scaffolding and create physical barriers that compartmentalize synapses and limit diffusion of neurotransmitters. In addition, glial cells express a variety of neurotransmitter receptors. Activation of glial receptors causes increases in intracellular calcium that in turn may lead to the secretion of factors that influence circuit function (Haydon and Nedergaard, 2015). In the developing brain, glial cells secrete factors that are important for neuronal differentiation, migration, axonal pathfinding and synaptogenesis (Clarke and Barres, 2013;Garrido, 2013;Chung et al., 2015). However, the mechanisms that are used for neuron-glial signaling in the developing CNS remain poorly defined.
There are three types of glial cells in the retina: Müller cells (MCs), astrocytes, and microglial cells. MCs, the principal glial cells of the retina, stretch radially across its thickness from the outer limiting membrane to the inner limiting membrane, sending processes throughout the inner and outer plexiform layers. MC processes envelop synapses of every neuron type, placing these cells in an ideal position to receive and respond to neuronal signals (Newman and Reichenbach, 1996;Newman, 2004Newman, , 2005Reichenbach and Bringmann, 2013). In adult retina, MCs maintain retinal homeostasis and integrity (Byrne et al., 2013), and more recently they have been implicated in delivery of retinol to replenish cone opsin chromophore (Xue et al., 2015). In zebrafish, MC ablation does not affect the laminar organization of the inner plexiform layer (IPL, Randlett et al., 2013) or the structure or organization of synapses in the outer plexiform layer (Williams et al., 2010), in sharp contrast to ablation of MCs in developing (Dubois-Dauphin et al., 2000) and adult mice (Byrne et al., 2013;Shen et al., 2012), which leads to dramatic rearrangements retinal structure (Byrne et al., 2013). However, unlike radial glial cells in the cortex and cerebellum, the role of MCs in retinal development has not been fully explored.
MCs are the last retinal cell types to mature, integrating into neuronal circuits during the first two postnatal weeks (Jadhav et al., 2009;Williams et al., 2010) when the inner retina is undergoing robust synaptogenesis. During this developmental period, the retina is spontaneously active, and ganglion cells (the output cells of the retina) exhibit depolarizations that spread laterally across the retina, a phenomenon termed retinal waves. These waves are mediated by circuits that change as the retina matures (Blankenship and Feller, 2010). From postnatal day 1 (P1) to P9, retinal waves are mediated by cholinergic circuits. As bipolar cells integrate into the circuit (P10-P14), the waves become mediated by glutamatergic signaling. At eye opening (P14), light activated signaling begins and retinal waves cease. Both cholinergic (Ford et al., 2012) and glutamatergic (Blankenship and Feller, 2010;Firl et al., 2013) retinal waves are associated with volume release of neurotransmitter, where the neurotransmitter is not restricted to the synaptic cleft, but spills out into the extracellular space, allowing activation of receptors on nearby ganglion cells. Hence, these waves provide a robust source of neurotransmitter that could mediate periodic activation of MCs during retinal development.
Here we investigated if MCs participate in retinal waves as they integrate into neuronal circuits. We simultaneously performed whole-cell recordings of retinal ganglion cells (RGCs) and two-photon imaging of calcium indicators or glutamate sensors expressed specifically in MCs in order to investigate if neurotransmitters released during retinal waves reliably induce calcium transients in MCs. In addition, we determined which receptors mediate these neuron-glia interactions. Our results indicate that MCs express a complement of receptors that allow detection of neurotransmitters released during retinal waves, resulting in parallel activation of this glial network prior to eye opening.

Results
Developing MCs exhibit calcium transients that correlate with retinal waves MCs undergo a dramatic morphological transformation during the second postnatal week, a time when there are robust retinal waves. At P7 ( Figure 1B, bottom), a MC displays a simple morphology eLife digest A structure at the back of the eye known as the retina is essential for vision. When light hits the retina, cells called neurons produce electrical signals that lead to the release of chemicals known as neurotransmitters. When these chemicals reach a neighboring neuron, a corresponding electrical signal is produced in this cell. In this way, information about what we can see is ultimately transmitted to the brain.
The retina also contains cells that are not neurons, such as Müller glial cells. These cells span the thickness of the retina, and appear well placed to interact with all of the different types of neuron in the retina. Other types of glial cells help the neural circuits in the brain to develop, but it is not clear whether Müller cells perform the same role in the developing retina.
Newborn mice do not open their eyes until around two weeks after they are born, during which time their retina continues to develop. The retinal neurons are already active throughout this period, and spontaneously generate periodic bursts of electrical activity that spreads across the developing retina in waves. Now, Rosa, Bos et al. show that in newborn mice, these waves also trigger brief, or 'transient', increases in the concentration of calcium ions inside Müller cells. Exactly how the Müller cells respond depends on which neurotransmitters are released during waves.
A few days before the mice open their eyes, the number of calcium transients in Müller cells decreases sharply. At the same time, neurons continue to spontaneously release waves of one particular neurotransmitter called glutamate. Rosa, Bos et al. used a sensor that showed where glutamate is found in the developing retina. This revealed that the decrease in calcium transients as the retina matures is due to less glutamate reaching the Müller cells.
These findings reveal that Müller cells are involved in the spontaneous electrical activity seen in the retina during a critical period of retinal development. The next challenge is to determine how this neuronal-glial cell signaling helps the retina to mature.  with its cell body residing in the inner nuclear layer (INL) and a thin, fiber-like process traversing the retina to form part of the inner and outer limiting membranes (green signal in GLAST:: tdTomato, middle panel, and bright punctas in GLASTCreER::GCaMP3, right panel). By P11 ( Figure 1B, middle), MCs have extended many lateral processes (yellow arrows) into the IPL and these continue to grow into the adult form where they closely appose synapses and participate in removing neurotransmitters from the extracellular space (Pow and Crook, 1996).
Like other types of glial cells, MCs in the mature retina exhibit calcium transients that occur spontaneously or are evoked by various stimuli (Metea and Newman, 2006;Reichenbach and Bringmann, 2013). However, MC calcium signaling during retinal development remains unexplored. Thus, we first investigated if MC calcium transients during early development coincided with retinal waves. We visualized MC transients via two-photon calcium imaging of whole-mount retinas from mice expressing a genetically encoded calcium indicator specifically in MCs (GLAST-CreER::ROSA26-lsl-GCaMP3 [GLAST-CreER::GCaMP3]; Figure 1A; see 'Materials and methods', [Paukert et al., 2014]). In the retina of GLAST-CreER mice, tamoxifen inducible Cre recombinase (CreER) is expressed by MCs ( Figure 1A). GLAST-CreER::GCaMP3 mice expressed sufficient GCaMP3 to detect MC calcium transients in the IPL as early as P7. Retinal waves were identified by the occurrence of compound postsynaptic excitatory currents in RGCs. Simultaneous two-photon imaging of MCs and whole-cell recordings from RGCs ( Figure 1C,D) showed periodic MC calcium transients in the stalks and lateral processes in the IPL (Figure 1-figure supplement 1, Video 1) that coincided with RGC compound postsynaptic excitatory currents (Blankenship et al., 2009). Since the stalks and processes of MCs exhibited similar calcium responses, we pooled their results together throughout this study. We detected no wave-evoked calcium transients in other parts of MCs outside the IPL (i.e., in their somata, data not shown). The percentage of regions of interest (ROIs, which correspond to compartments of individual MCs, see Figure 1-figure supplement 1) that responded to a wave (termed responsive MCs) was high at P7 (42 ± 10.2%, 1326 ROIs from 11 retinas) and at P9 (48 ± 4.6%, 3027 ROIs from 14 retinas), but significantly lower at P11 (13 ± 2.2%, 872 ROIs from 6 retinas, Figure 1E). As MCs express a variety of neurotransmitter receptors, including glutamatergic and cholinergic receptors (Wakakura and Yamamoto, 1994;Belmonte et al., 2000), MC calcium transients at different ages could be evoked by different neurotransmitters released during retinal waves. Thus, we next explored which transmitters modulated neuron-MC signaling at different developmental ages.

MC calcium transients correlated with cholinergic retinal waves are mediated by muscarinic acetylcholine receptors
Our primary hypothesis is that MC calcium transients are induced by neurotransmitters released from amacrine and bipolar cells (the interneurons of the retina) during retinal waves. To assess which neurotransmitters could elicit MC calcium transients during development, we first imaged MC calcium signals in the IPL in response to periodic, focal application of agonists that could be potentially involved in the neuron-glia interaction during P7 cholinergic waves (Figure 2A-E). Control application of extracellular solution (artificial cerebrospinal fluid [ACSF]) did not evoke a MC response, indicating that the pressure injection itself did not evoke calcium transients through mechanical stimulation ( Figure 2C). When adenosine tri-phosphate (ATP, 1 mM) was applied, robust calcium transients were induced that were inhibited by the P2 receptor blocker suramin (100 μM; Figure 2D), as seen previously in the adult retina (Uckermann et al., 2002;Newman, 2004;Metea and Newman, 2006). MCs also responded to acetylcholine (ACh, 1 mM; Figure 2B,C; Video 2), as described in cortical   (Takata et al., 2011). These AChevoked MC calcium transients were reduced by the muscarinic ACh receptor antagonist atropine (50 μM; Figure 2E). Similar ACh-and ATPevoked MC calcium transients were also observed at P9 (during the transition from cholinergic to glutamatergic waves) and at P11 (during glutamatergic waves), indicating that MCs express multiple neurotransmitter receptors prior to eye opening ( Figure 2D,E).
To explore the contribution of ACh and ATP to MC calcium transients that occur during P7 cholinergic waves, simultaneous two-photon laser scanning microscopy and whole-cell RGC recordings were performed. MC calcium transients ( Figure 2F, top black traces) were correlated with RGC compound excitatory postsynaptic currents (EPSCs) that are known to be mediated by the volume release of ACh and the activation of neuronal nicotinic ACh receptors (AChRs) on RGCs ( Figure 2F, bottom grey traces) (Ford et al., 2012). Interestingly, this ACh is released by cholinergic amacrine cells that also release ATP (Duarte et al., 1999). Thus, we tested the effect of AChR and ATP receptor blockers on wave-induced MC calcium transients. Although the wave-induced transients were blocked by the muscarinic ACh receptor (mAChR) antagonist atropine (50 μM; Figure 2G), they were not blocked by the ATP receptor antagonist suramin (100 μM; Figure 2H). Therefore, we conclude that at P7, MC calcium transients are evoked by volume release of ACh, but not by ATP, and these transients are mediated by mAChRs on MCs.
MC calcium transients correlated with glutamatergic retinal waves are limited by the extent of glutamate spillover and are mediated by AMPA receptors At P9, circuits mediating retinal waves change from cholinergic to glutamatergic (Blankenship and Feller, 2010), and glutamatergic synapses between bipolar cells and their postsynaptic targets first appear (Hoon et al., 2014). Thus, at P9 cholinergic and/or glutamatergic transmission could activate MC calcium transients. To determine the involvement of cholinergic transmission in MC signaling at this stage of development, we examined their response to focal application of ACh. Short applications of ACh induced calcium transients that were blocked by the mAChR blocker atropine at P9 ( Figure 2E), but bath application of atropine did not change the percentage of wave-induced MC calcium transients (57 ± 12.6% in control and 60.2 ± 7.8% in atropine; p = 0.54, n = 4). These results indicate that cholinergic signaling does not mediate neuron-MC signaling at this age. We then determined the role of glutamate in triggering these responses. Application of exogenous L-glutamate reliably evoked MC calcium transients (Figure 3A-C; Video 3). These responses were not blocked by atropine, ruling out the possibility that this agonist acted indirectly by stimulating release of ACh from amacrine cells. L-glutamate-evoked MC calcium transients were significantly reduced by a cocktail of ionotropic glutamate receptor antagonists (iGluR: 20 μM AMPA/kainate   Figure 3D), but were only slightly reduced by a mixture of metabotropic glutamate receptor antagonists (mGluR: 8 μM LY341495, 100 μM MCPG, 100 μM MTEP; Figure 3D). Blocking AMPA receptors produced substantial but incomplete inhibition of the L-glutamate-evoked response, whereas blocking NMDA receptors had no significant effect. These results suggest that glutamate elicits MC calcium transients primarily through activation of AMPA receptors, although a modest contribution of mGluRs is also observed. Notably, the reduction of L-glutamate-evoked calcium transients by glutamate receptor blockers occurred at all ages, not just during the period when glutamatergic waves predominate, indicating that MCs express glutamate receptors throughout this period of development.
To study MC calcium transients evoked by glutamatergic waves, we combined two-photon laser scanning microscopy with whole-cell recordings from RGCs at P9 and P11. At P9, wave-induced MC calcium transients were correlated with compound EPSCs in RGCs ( Figure 3E,F) that are mediated by release of glutamate from bipolar cells and by activation of ionotropic glutamate receptors on RGCs (Bansal et al., 2000;Wong et al., 2000;Zhou and Zhao, 2000). However, at P11, wave-induced MC calcium transients were dramatically reduced ( Figures 1E, 3G,H) although robust glutamate spillover occurs at this age (Blankenship et al., 2009;Firl et al., 2013) and exogenous glutamate induced MC calcium transients ( Figure 3B-D).
We hypothesize that this reduction of MC activity is a consequence of the upregulation of glutamate transporters at P11 (Pow and Barnett, 2000;Pannicke et al., 2002), which may limit the amount of glutamate that reaches MC membranes to induce calcium transients. To test this hypothesis, we blocked glutamate transporters by bath applying DL-TBOA (25 μM), an agent previously shown to increase extracellular glutamate levels (Blankenship et al., 2009;Firl et al., 2013). At P11, DL-TBOA drastically increased the percentage of responsive MCs from 13 ± 2.2% to 61 ± 4.3% ( Figure 3G,H; 872 ROIs in control and 1342 ROIs in DL-TBOA from 6 retinas). Interestingly, at P9, DL-TBOA did not significantly change the percentage of responsive MCs, increasing only slightly from 48 ± 4.5% to 58 ± 10.0% ( Figure 3E,F; 3027 ROIs in control and 3549 ROIs in DL-TBOA from 8 retinas). We conclude that at ages earlier than P11, transporters do not strongly regulate glutamate spillover, but by P11, they mediate significant glutamate uptake, decreasing the amount of glutamate that reaches MC receptors. We expressed the membrane-bound intensitybased glutamate sensing fluorescent reporter iGluSnFR  either in MCs or neurons  to test directly if extracellular glutamate was reaching the membrane of MCs during retinal waves. The iGluSnFR signal is proportional to the amount of glutamate reaching the membrane ; thus, the lack of fluorescence in the INL where MC cell bodies are located indicates that during early development there is little glutamate release outside of the IPL.
To detect glutamate signals during retinal waves, we simultaneously performed two-photon imaging of iGluSnFR signals in the IPL and whole-cell voltage clamp recordings from RGCs    At both P9 and P11, the processes of neurons that expressed iGluSnFR exhibited large glutamate transients that propagated throughout the field of view with each retinal wave ( Figure 4A,C; 160 ROIs from 5 retinas at P9 and 160 ROIs from 5 retinas at P11; Video 4), consistent with the key role played by glutamate in inducing retinal waves at these ages. In contrast, although many MCs expressing iGluSnFR displayed glutamate transients during P9 waves (38 ± 4%, 1023 ROIs from 5 retinas, Figure 4B,C), the percentage of responsive MCs decreased dramatically by P11 (6 ± 2%, 269 ROIs from 5 retinas at P11; Figure 4B,C). Inhibiting glutamate transporters with DL-TBOA (25 μM) increased the percentage of responsive MCs at P11 (31 ± 7%, 822 ROIs from 5 retinas; Video 5) and reduced the inter-event intervals for MC calcium and glutamate transients ( Figure 4D; Figure 4-source data 1), indicating that glutamate uptake limits spillover-induced activation of MCs at this age. DL-TBOA exerted a minimal effect on MC calcium and glutamate transients at P9 ( Figure 4B-D). These findings corroborate the hypothesis that the lower expression of glutamate transporters at P9 enables robust entrainment of MCs during retinal waves.
Several lines of evidence support the hypothesis that MC calcium transients depend upon the direct activation of MC glutamate receptors during the period of glutamatergic retinal waves. First, MCs are known to express AMPARs (Wakakura and Yamamoto, 1994;Sullivan and Miller, 2010). Second, glutamate-induced MC calcium transients depend upon the activation of ionotropic AMPARs ( Figure 3D). Third, the frequencies of glutamate and calcium transients in MCs are indistinguishable, indicating that MC calcium transients are driven by glutamate reaching the MC membranes ( Figure 4D; Figure 4-source data 1). Establishing that wave-induced MC calcium transients are mediated by ionotropic glutamate receptors on MCs requires selective inhibition of these receptors; however, this approach is complicated by the fact that depolarization of RGCs during retinal waves also depends on AMPAR activation. Since the AMPAR antagonist DNQX may have an effect on glutamate release from bipolar cells, we tried to elucidate its effects in this circuit. In the presence of DNQX, we simultaneously monitored EPSCs from RGCs with whole-cell recordings and glutamate release from bipolar cells with neuronal iGluSnFR imaging ( Figure 5A,B). As described previously (Blankenship et al., 2009), DNQX (20 μM) significantly decreased the amplitude of the postsynaptic current in RGCs. However, it did not change the amount of wave-induced glutamate release from bipolar cells monitored using iGluSnFR ( Figure 5A). With this knowledge, we used DNQX to determine the role of AMPARs in wave-induced MC calcium transients. These experiments were conducted in the presence of DL-TBOA to maximize the Our results strongly suggest that glutamate acts directly on MC AMPARs to increase intracellular calcium. However, it will be necessary to selectively delete AMPARs from MCs to establish their contribution to MC calcium signaling during this period of development.
Previous studies have shown that mGluRs contribute to neuron-astrocyte signaling in the brain, particularly during early development (Grosche et al., 1999;Kirischuk et al., 1999).
To determine the extent that mGluRs are involved in wave-induced MC calcium transients in the retina we performed two experiments. First, we conducted simultaneous whole-cell recordings from RGCs and two-photon calcium imaging of MC activity in the presence of mGluR antagonists (8 μM LY341495, 100 μM MCPG, 100 μM MTEP). In these conditions, there was a small, but not significant, decrease in the percentage of responsive MCs at P9 ( Figure 6D), consistent with the contribution of mGluRs to MC calcium transients evoked by exogenous L-glutamate ( Figure 3B-D). Second, we imaged the spontaneous activity of MCs in retinas isolated from mice that lack IP 3 R2 (IP 3 R2-KO mice), which has been shown to be required for metabotropic receptor-induced calcium signaling in adult MCs (Lipp et al., 2009), using a fluorescent calcium dye ( Figure 6A,B). The density of MC labeling was not dramatically different between GCaMP3 expressing mice vs MCs loaded with organic dyes, although this was not compared quantitatively. Wave-like calcium transients persisted in IP 3 R2-KO MCs and a similar percentage of MCs participated in these events ( Figure 6C,D) indicating that IP 3 -dependent calcium signaling contributes minimally to MC calcium transients during retinal waves. Together, these findings suggest that volume release of glutamate during retinal waves evokes MC calcium transients primarily by activating AMPA receptors on MCs.

Discussion
In this study we elucidated the spatio-temporal characteristics and molecular pathways by which neurons and MCs communicate in the developing retina. We found that retinal waves were correlated with calcium transients in a large proportion of MCs, demonstrating that spontaneous activity encompasses both neuronal and glial networks at this age. Our physiological results indicate that MCs express multiple neurotransmitter receptor types throughout this period, allowing neuronal control of MC activity to be sustained as neuronal synchronization shifts from being dependent on cholinergic to glutamatergic signaling. The proportion of MCs that responded to waves with calcium transients decreased significantly from P9 to P11, despite the presence of retinal waves. This age-dependent decline in MC activation during retinal waves was caused by an increase in glutamate transporter expression, which restricted glutamate spillover from nearby synapses limiting activation of MC AMPARs. These data indicate that MCs in the developing retina use multiple signaling pathways to participate in spontaneous, correlated activity with neurons during a crucial period of retinal development.
Many different signaling mechanisms have been implicated in inducing calcium transients in astroglial cells throughout the peripheral and central nervous system Haydon and Nedergaard, 2015;Nimmerjahn and Bergles, 2015). The robustness and mechanisms of signaling vary tremendously, even at the same synapses, with age, preparation, and state of arousal. Our data indicate that MCs in the retina prior to eye opening are capable of responding to different neurotransmitters with increases in intracellular calcium (Figures 2A, 3A). Neurotransmitter-evoked calcium transients were observed in the stalks and lateral processes of MCs that transverse the IPL during a time when there is robust synaptogenesis. These multiple receptors allow MCs to detect neurotransmitter released during retinal waves whether it is ACh or glutamate, thereby extending the developmental period over which neuron-glia signaling occurs.
The neuron-glia signaling we observed between neurons and MCs in the retina during glutamatergic waves is similar to signaling that occurs in Bergmann glial cells (BGs), a type of radial astroglial cell found in the cerebellum. BGs express calcium permeable AMPARs that are activated by glutamate released at parallel and climbing fiber synapses (Bergles et al., 1997;Clark and Barbour, 1997;Piet and Jahr, 2007), raising the possibility that AMPAR expression may be a conserved properties of radial-type astroglial cells. Although the role of AMPARs in BG activity is still controversial (Müller et al., 1996;Iino et al., 2001;Nimmerjahn et al., 2009), in vivo studies indicate that calcium signaling in BGs during locomotion depends on the activation of AMPARs (Nimmerjahn et al., 2009) and disruption of this signaling (Iino et al., 2001;Saab et al., 2012), results in prolonged synaptic events that are associated with altered motor behavior and motor learning. Moreover, the mechanism of interaction between neurons and MCs we observed during cholinergic retinal waves is similar to previous findings in visual cortex and hippocampus, where cholinergic neuron stimulation leads to astrocyte calcium transients via activation of muscarinic acetylcholine receptors and require IP3R2 (Chen et al., 2012;Navarrete et al., 2012).
Whether neuronal activity during retinal waves guides the development of MCs is still unclear. MCs exhibit dramatic morphological changes during the period from P7-P11 as their fine processes grow laterally into the IPL and ensheath synapses. Ensheathment of synapses by astrocytes has been shown to promote functional isolation of synaptic signaling by creating a barrier to diffusion of neurotransmitters and by placing neurotransmitter transporters near sites of release. It is possible that ablation of AMPARs in MCs would result in retraction of the lateral processes away from synapses, as observed in BGs after inhibition of calcium permeable AMPARs (Iino et al., 2001) or genetic ablation of GluA1 and GluA4 (Saab et al., 2012). Our findings in the developing retina differ from studies performed in the adult retina, in which glutamate does not evoke calcium transients in MCs (Newman and Zahs, 1997;Newman, 2005), suggesting that glutamaterigic signaling through AMPARs in these cells may be limited to the early stages of synapse formation and refinement.
Astrocytes have also been shown to play critical roles in promoting the refinement and maturation of neural circuits (Stevens, 2008;Clarke and Barres, 2013;Chung et al., 2015) by secreting synaptogenic factors, such as thrombospondin, hevin, protocadherins, BDNF, and TGFbeta, which are critical for synapse formation between RGCs and their central targets in the brain (reviewed in Chung and Barres, 2012;Clarke and Barres, 2013). In addition, astrocytes play an important role in refining nascent circuits in the visual system by engulfing unused or inappropriate synapses (Chung et al., 2015). Our findings indicate that robust signaling occurs between retinal neurons and MCs between the ages P7 and P11, a time period when there is robust synaptogenesis in the retina (Hoon et al., 2014). The close lineage relationship and functional similarities between astrocytes and MCs increase the possibility that MCs perform similar roles in the retina, particularly in regions such as the IPL where they are the sole astroglial cell.
Previous studies have shown that changes in intracellular calcium can trigger the release of socalled gliotransmitters such as ATP, glutamate and D-serine from astrocytes that can influence the activity of surrounding neurons (Araque et al., 2014;Newman, 2015). In the retina, activation of AMPARs promotes the release of D-serine, a co-agonist of NMDA receptors in ganglion cells, that has been shown to play an important role in synaptic plasticity (Stevens, 2008;Sullivan and Miller, 2010), raising the possibility that MC activity induced during retinal waves participates in synaptic refinement through release of gliotransmitters and/or synaptogenic factors. Selective manipulation of MC calcium signaling during this period of remarkable refinement will help to further define their roles in the development of distinct retinal circuits.

Materials and methods Animals
GLASTCreER::GCAMP3 mice were generated by cross breeding GLAST-CreER BAC transgenic mice [Tg(Slc1a3-cre/ERT)1Nat/J] (de Melo et al., 2012) to ROSA26-lsl-GCaMP3 reporter mice (Paukert et al., 2014). GLASTCreER::tdTomato mice were generated by crossing GLASTCreERBAC mice to mice that ubiquitously express tdTomato preceded by a loxP-flanked stop cassette (B6;129S6-Gt (ROSA)26Sor tm9(CAG-tdTomato)Hze /J, from The Jackson Laboratory, Bar Harbor, ME). The gene targeting and generation of IP 3 R2 knock-out mice have been previously described (Li et al., 2005). All animal procedures were approved by the University of California (UC) Berkeley and Johns Hopkins University's Animal Care and use Committees and conformed to the NIH Guide for the Care and Use of Laboratory Animals, the Public Health Service Policy, and the SFN Policy on the Use of Animals in Neuroscience Research.
RGC dendritic stratification was visualized by including 20 μM Alexa Fluor 594 (Invitrogen, Grand Island, NY) in the intracellular solution. The dendritic morphology of dye-injected RGCs was reconstructed by two-photon imaging with the laser tuned to 780 nm. Images (RGCs and whole retina) were acquired at z intervals of 0.5 μm using a 60× objective (Olympus 60×, 1 NA, LUMPlanFLN). Images were later reconstructed from image stacks with ImageJ.

Focal agonist stimulation
Direct activation of receptors on MCs was accomplished by repeated short applications (100 ms) of L-glutamate (1 mM, Sigma-Aldrich, St Louis, MO), acetylcholine (ACh, 1 mM, Sigma-Aldrich, St Louis, MO), ATP (1 mM, Tocris Bioscience, Minneapolis, MN), or ACSF to the IPL via a pipette that was similar to the recording pipette and a Picospritzer (PV830 Pneumatic PicoPump, World Precisions Instruments, Sarasota, FL) set at 20-30 psi.

Image analysis
Videos were analyzed offline using SARFIA (freely available on http://www.igorexchange.com/project/ SARFIA), a suite of macros running in Igor Pro (Wavemetrics, Portland, OR) (Dorostkar et al., 2010) Prior to analysis, images were registered to correct movements in the X and Y directions. Videos were rejected if the plane of focus altered significantly during imaging acquisition. For GLASTCreER:: GCaMP3 and ShH10-CMV-iGluSnFR recordings, ROIs containing both stalks and lateral processes of MCs were chosen using a filtering algorithm based on a Laplace operator and segmented by applying a threshold, as described in detail in (Dorostkar et al., 2010). This algorithm defined most or all of the ROIs that an experienced observer would recognize by eye. For AAV9-2YF-hSynapsin1-iGluSnFR recordings, the entire field of view (256 × 256 pixels) was divided into 16 identical squares and each square was considered one ROI. In both cases, individual ROI responses were then normalized as the relative change in fluorescence (ΔF/F), smoothed by binomial Gaussian filtering, and analyzed to detect activity using custom-made scripts based on a first derivate detection algorithm. A threshold set at ∼2 times (when imaging at 0.74 Hz) or at ∼4 times (when imaging at 1.7 Hz) the standard deviation of the time derivative trace was used to detect changes in fluorescence within the ROIs. The reliability of this algorithm to detect calcium and glutamate activity was first tested by comparing the results with manual activity detection. For the experiments using focal application of agonists, a threshold set at ∼2-3 times the standard deviation of the derivate trace was used to detect activity. In this paper, the fluorescent intensity of ROIs is reported as the average intensity across all pixels within its area. Fluorescent responses are reported as normalized increases as follows: where F is the instantaneous fluorescence induced by a spontaneous activity or by a focal application of the agonist and F o is the baseline fluorescence.

Statistical analysis
Group measurements are expressed as mean ±SD. We used t tests to compare two groups and oneway ANOVA to compare more than two groups. The level of significance was set at p < 0.05.