Visual Circuit Development Requires Patterned Activity Mediated by Retinal Acetylcholine Receptors

SUMMARY

The elaboration of nascent synaptic connections into highly ordered neural circuits is an integral feature of the developing vertebrate nervous system. In sensory systems, patterned spontaneous activity before the onset of sensation is thought to influence this process, but this conclusion remains controversial largely due to the inherent difficulty recording neural activity in early development. Here, we describe novel genetic and pharmacological manipulations of spontaneous retinal activity, assayed in vivo, that demonstrate a causal link between retinal waves and visual circuit refinement. We also report a de-coupling of downstream activity in retinorecipient regions of the developing brain after retinal wave disruption. Significantly, we show that the spatiotemporal characteristics of retinal waves affect the development of specific visual circuits. These results conclusively establish retinal waves as necessary and instructive for circuit refinement in the developing nervous system and reveal how neural circuits adjust to altered patterns of activity prior to experience.

INTRODUCTION

Neural circuit development is a complex process that integrates cell differentiation, migration, neurite outgrowth, axon pathfinding, synapse formation, elaboration, and refinement in order to equip the nervous system for early-life behavior. A unique feature of this process is the interaction between neuronal activity and synapse formation, which creates the opportunity for neural network dynamics to contribute to fundamental aspects of synapse and circuit development. In some respects, this is a precocious example of “learning” in which neuronal activity causes circuit rewiring through synaptic plasticity. Here, we describe experiments that examine the link between initial in vivo activity patterns and neural circuit refinement in the mammalian visual system, and demonstrate that early activity patterns are not only necessary and instructive for initial circuit development, but are optimized for the generation of distinct circuit features.

In contrast to lower vertebrates and insects, in which the majority of initial circuit development is believed to be dependent on activity-independent molecular factors, model circuits in a variety of mammalian systems refine using both molecular and activity dependent factors (Sperry, 1963; Sanes and Zipursky, 2010; Kirkby, Sack et al. 2013). This process is best characterized in the visual system (Huberman et al. 2008; Cang and Feldheim, 2013; Ackman and Crair, 2013), where classic experiments suggest the necessity of neuronal activity for normal circuit development (Hubel and Wiesel, 1970; Shatz and Stryker, 1988), while more recent discoveries identify a host of molecular factors that together play important roles in neurite targeting, elaboration, and refinement (Cang and Feldheim, 2013). However, the difficulty inherent in recording and manipulating activity in developing animals and the continued presence of some normally targeted axons even after the removal of multiple guidance cues (Frisen et al., 1998; Pfeiffenberger et al., 2006) leaves room for debate over the relative contributions of activity and molecular guidance in this process.

In the visual system, classic experiments demonstrate that visual experience fundamentally shapes circuit development during a critical period of development (Hubel and Wiesel, 1970; Espinosa and Stryker, 2012). However, essential features of developing mammalian neural circuits emerge before sensory experience is possible (Rakic, 1976; Horton and Hocking, 1996; Crair et al. 2001; Crowley and Katz, 2000), or even in the absence of vision (Crair et al., 1998; White et al., 2001). Moreover, stereotypical features of central visual circuits endure following the early removal of both eyes, implying that retinal input is not necessary for the development of some circuits (Ruthazer and Stryker, 1996; Crowley and Katz, 1999). These apparently contradictory results cast doubt on the necessity and relevance of neural activity in the early emergence of visual circuits, and led to the theory that spontaneous activity in the developing retina and visual system could drive circuit refinement prior to sensory experience (Galli and Maffei, 1988; Shatz and Stryker, 1988). Pioneering experiments demonstrated that perinatal retinas do indeed have correlated spontaneous bursts of activity (Galli and Maffei, 1988) that are organized into large-scale, seemingly random patterns termed “retinal waves” (Meister et al., 1991). Subsequent research established that this patterned retinal activity propagates throughout the developing visual system (Weliky and Katz, 1999; Ackman et al., 2012).

Experiments designed to determine the degree to which visual circuit development depends on retinal waves have proven less conclusive. Results from chronic in vivo pharmacological manipulations that are known to either disrupt or agonize retinal waves in vitro (Penn et al., 1998; Stellwagen and Shatz, 2002), or eliminate activity conductance through the optic nerve (O’Leary et al., 1986), generally support a link between retinal waves and circuit refinement. However, the interpretation of these experiments is complicated by the variable and cell-type specific influence of pharmacological manipulations (Sun, Speer et al., 2008), the potential systemic and/or neurotoxic effects of chronic injections (O’Leary, Crespo et al., 1986; Ackman et al., 2012; Shatz and Stryker, 1988; but also see Penn et al., 1998), and only limited knowledge about the retinal or downstream activity patterns following these manipulations in vivo. Intriguingly, experiments employing a toxin that kills starburst amacrine cells, the retinal cell type known to nucleate early postnatal retinal waves (Zhou, 1998; Zheng et al, 2006), found de-correlation of patterned activity in vitro without any effect on visual circuit development (Huberman et al., 2003). This work implied that only the existence of retinal ganglion cell bursting activity, but not patterned retinal waves, was necessary for circuit refinement.

Genetic models that disrupt spontaneous retinal activity circumvent many of the experimental complications of pharmacological manipulations, but come with their own caveats. The most widely utilized knockout model for retinal waves are β2^−/− mice, which lack expression of the β2-subunit of the nicotinic acetylcholine receptor (β2-nAChR) (Bansal et al., 2000). In vitro pharmacological manipulations demonstrate that spontaneous retinal waves during the first week after birth rely on the expression of β2-nAChRs by starburst amacrine cells (Feller et al., 1996; Zhou, 1998), and initial reports suggested that β2^−/− mice completely lack retinal waves (Bansal et al., 2000; McLaughlin et al., 2003). These findings, along with the relative ease and reproducibility of the β2^−/− model in comparison to chronic pharmacological injections, resulted in a host of reports broadly linking retinal waves and β2^−/− mice to altered retinotopy, eye-specific segregation, visuotopic map alignment, and functional deficits throughout the visual system (McLaughlin et al., 2003; Rossi et al., 2001; Chandrasekaran et al., 2005; Mrsic-Flogel et al., 2005; Triplett et al., 2009; Grubb and Thompson, 2004; Cang et al. 2005; Stafford et al., 2009; Shah and Crair, 2008; Wang et al., 2009).

However, subsequent reports revealed that retinal waves actually persisted in β2^−/− mice under some recording conditions, perhaps even with greater frequency, though their spatiotemporal properties were usually altered (Sun, Speer et al., 2008; Sun, Warland et al., 2008; Stafford et al., 2009). This suggests that disrupted spatiotemporal patterns of waves and not a decrease in wave frequency are the likely cause of circuit refinement defects in β2^−/− mice (Godfrey et al., 2009; Stafford et al., 2009). Further complicating matters, experiments employing β2^−/− mice suffer from the caveats typical of whole-animal knockouts, since β2-nAChRs are normally expressed throughout the brain and body, and glutamatergic synapse development is disrupted in the CNS of β2^−/− mice (Shah and Crair, 2008; Lozada et al., 2012) completely apart from their likely retinal wave deficits. These and other experimental ambiguities cast some doubt on the existence, form, and necessity of retinal waves for visual circuit development (Chalupa, 2009; Feller, 2009).

We report here on experiments examining in vivo the causal link between retinal waves and visual circuit refinement using a combination of imaging, anatomical circuit tracing, and pharmacological manipulations in β2^−/− and novel conditional β2-nAChRs mutant mice. Our in vivo recordings demonstrate that in whole-animal β2^−/− mice there is a marked decrease in patterned spontaneous retinal activity and dramatically disrupted waves. We also observe surprisingly strong correlations in spontaneous activity between the eyes and, remarkably, a de-correlation of presynaptic (retinal) and postsynaptic (collicular) activity in β2^−/− mice. Using conditional β2-nAChRs mutant mice, we further show that retinal, but not collicular β2-nAChRs are necessary for both retinal waves and visual circuit refinement. Moreover, deletion of β2-nAChRs from restricted zones in the retina eliminates retinal waves and disrupts visual circuit refinement, but regions of the same retina with maintained expression of β2-nAChRs continue to wave and exhibit normal circuit refinement. Finally, we demonstrate that chronic pharmacological rescue of retinal wave frequency partially restores visual circuit development in β2^−/− mice. Taken together, these results substantiate a strong causal link between β2-nAChRs in the retina, spontaneous retinal waves, and visual circuit refinement in the developing lateral geniculate nucleus (dLGN) and superior colliculus (SC) of mice.

RESULTS

Retinal Ganglion Cell Spontaneous Activity in β2^−/− Mices

We examined spontaneous activity in retinal ganglion cells in vivo during the first postnatal week (P3–P7) by loading retinal ganglion cells (RGCs) with an exogenous (Calcium Green-1 Dextran) or genetically encoded (GCaMP3/6) fluorescent calcium indicator. Retinal waves were imaged in topographically mapped RGC axon arbors in the superior colliculus (Fig. 1). This approach reveals frequent waves of propagating spontaneous activity that gradually encompasses much of the SC (Ackman et al. 2012) (Fig. 1A–C). In mice lacking the β2 subunit of the nicotinic acetylcholine receptor (Chrnb2) throughout the brain and body (β2^−/− mice), correlated RGC activity was typically present, but much less frequent than in control (WT and β2^+/−) mice and with noticeably altered spatiotemporal properties (1.00 ± 0.06 waves/min in n= 27 WT SC hemispheres; 0.34 ± 0.05 waves/min in n= 56 β2^−/− hemispheres; 1.23 ± 0.08 waves/min in n= 8 β2^+/− hemispheres; mean ± S.E.M; p < 0.001 for the difference between WT and β2^−/−; Movie S1, Fig. 1D, Fig. 2). The frequency of spontaneous activity in β2^−/− mice was particularly low at the beginning of the first week after birth (Fig. 1E), when visual circuit refinement is most robust (Dhande et al., 2011), but then gradually increased to levels that were indistinguishable from control wave frequency one week after birth (P7). Remarkably, retinal spontaneous activity in β2^−/− mice was notably more correlated between the eyes than in control mice (Fig. 1C, F), as measured by comparing the time difference between waves in the two hemispheres (Inter-retina wave interval = 15.6s ± 12.1 in n = 16 WT mice; 3.7s ± 36.0 in n = 31 β2^−/− mice; 12.3s ± 8.4 in n = 8 β2^+/− mice; median ± median absolute deviation; p < 0.001 for K.S. test between probability distributions in WT and β2^−/− mice; Fig. 1F, Supplementary Experimental Procedures). A Monte-Carlo analysis of inter-hemispheric wave interval confirmed that β2^−/− mice were much more likely to have significantly coincident waves in the two eyes than littermate control mice (28/38 ten-minute epochs significantly correlated in β2^−/−; 1/16 epochs were correlated in WT; 0/6 epochs correlated in β2^+/− mice; see Fig. S1 and Supplementary Experimental Procedures). This surprising result may contribute to the disruption of eye-specific segregation observed in β2^−/− mice, as computational (Butts et al., 2007) and experimental (Zhang et al. 2011) results suggest that random between-eye wave intervals encourage the refinement of these circuits.

Figure 1. Infrequent spontaneous retinal activity in vivo in β2^−/− mice.

A. Retinal wave imaging schematic (top) and retinotopy cartoon (bottom). Retinal waves are imaged in retinal ganglion cell axon arbors in the upper layers of the SC, which are topographically mapped between the retina and superior colliculus (bottom). Retinal waves are imaged in RGC axon arbors in vivo through a craniotomy over the superior colliculus (SC). Calcium activity in RGC axons in the SC is measured either through direct injection of Calcium Green-1 Dextran into the retina or through the retina-specific expression of GCaMP3. B. Example single-wave montages from P3 WT mouse (top) and P3 β2^−/− mouse (bottom). Grey scale images on left show craniotomy over right and left hemispheres of SC. All movies are acquired at 5–10 Hz. Movie frames shown in montage at 2 second intervals (top) or 1 second intervals (bottom), scale bar = 200 μm. White arrows show onset and propagation of retinal wave front. Clear propagating retinal wave front is typical in WT mice (top), but difficult to discern in β2^−/− mice (bottom). C. Example raster plots of 5 minute recordings from P3 WT (left) and β2^−/− (right) mice. Each row in the raster corresponds to one 10 × 10 μm region of interest in the indicated hemisphere. D. Spontaneous retinal activity is much less frequent in β2^−/− mice than WT or β2^+/− littermate controls (for this and all subsequent figures, box plots are aggregated by SC hemisphere, box edges are first and third quartile, whiskers are 1.5 times the interquartile range, midline is median value, * - p<0.05, ** - p<0.01, *** - p<0.001). E. Spontaneous activity by genotype and age during the first postnatal week. Coordinated activity is much less frequent in β2^−/− mice than littermate controls. F. Waves are much more likely to occur at the same time between the two retinas in β2^−/− mice relative to littermate controls irrespective of wave frequency (results of frequency-correcting Monte Carlo analysis described in text, Figure S1). Plot shows cumulative probability distribution of the inter-retina wave interval (time interval between a wave in one retina compared to the other retina), inset box plot shows median. See also Figure S1.

Figure 2. Altered spontaneous retinal activity in β2^−/− mice.

A. Example single retinal wave montage from a β2^−/− mouse (scale-bar = 200 μm, time interval = 1s, age = P6). B. Analysis of wave duration (in seconds) in WT, β2^−/−, and β2^+/− mice reveals that waves are much more brief in β2^−/− relative to controls. C. Wave speed (μm/s) is also much faster in β2^−/− relative to controls. D. RGC activity is correlated over a much larger area of the retina in β2^−/− mice relative to controls. This is quantified by comparing the size of the wave at its peak relative to the overall spatial extent of the wave throughout its duration (see Experimental Methods). E. Wave directionality plots in β2^−/− and WT mice, binned by overall wave duration. β2^−/− retinal waves show greater directional preference than WT waves and are strongly biased in the anterior-posterior direction. F. Quantification of wave direction bias index (see Experimental Methods) shows that waves in β2^−/− mice have a stronger direction bias than littermate controls. See also Figure S2.

In addition to a frank reduction in retinal wave frequency (Fig. 1), a wide variety of spontaneous activity properties differed between β2^−/− and control mice (Fig. 2; Table 1). In general, correlated spontaneous activity in RGCs of β2^−/− mice was brief, of low amplitude, and lacked a well-defined wave front in comparison to littermate control mice (Fig. 2B, C, Fig. S2). Retinal wave size was also larger in β2^−/− mice relative to WT mice (Table 1; Fig. S2), and produced much larger correlations across the retina relative to controls (Area of wave at peak/total wave area = 0.32 ± 0.09 in n = 16 WT hemispheres; 0.51 ± 0.10 in n = 31 β2^−/− hemispheres; p < 0.001; Fig. 2D; see Experimental Procedures). The brief, large, and correlated activity observed in β2^−/− mice was largely eliminated following enucleation or application of the Gap junction antagonist Meclofenamic Acid (4–20 mM) to the eye (Fig. S2), consistent with in vitro experiments in perinatal WT and β2^−/− mice (Sun, Warland et al., 2008; Blankenship et al., 2011; Stacy et al., 2005). It was previously reported (Stafford et al., 2009) using in vitro recordings that retinal waves in β2^−/− mice lacked the directional bias that is observed in wild type mice both in vitro (Stafford et al., 2009) and in vivo (Ackman et al., 2012). Interestingly, we observed that retinal waves in vivo showed an even greater directional bias in β2^−/− mice than in WT mice (Direction Bias Index 0.31 ± 0.04 in n = 16 WT hemispheres; 0.61 ± 0.05 in n = 29 β2^−/− hemispheres; 0.35 ± 0.05 in n = 8 β2^+/− hemispheres; mean ± S.E.M; p < 0.001 for difference between β2^−/− and either WT or β2^+/− mice; Fig. 2E,F). Overall, these experiments show that correlated retinal activity is much less frequent in β2^−/− mice, but partially recovers by the end of the first postnatal week. At the same time, many properties of the residual spontaneous retinal activity in β2^−/− mice are dramatically different from control mice, with the fast, brief, directionally biased, and more correlated waves in β2^−/− mice potentially all contributing to disruptions in visual circuit refinement.

Table 1. Values and statistics for wave properties across genotypes and conditions.

Measurement descriptions for each row are located in the first column and the genotypes and conditions follow to the right. Measurements are aggregated by hemisphere and reported as means (+/− standard error of the mean). Statistical significance was determined by using one or two-way analysis of variance (ANOVA) along with Tukey’s honest significant difference (HSD) post-hoc test for reported p-values. Asterisks denote significant difference from the most comparable wild-type measurement. Pound signs denote significant difference between most comparable retinal and superior colliculus wave recordings (pre- versus postsynaptic wave differences). Dollar signs denote significant difference in β2^−/− wave properties after retinal injection of CPT-cAMP.

Measurement	WT	β2−/−	β2+/−	En1-Cre+ β2fl/−	Pax6α-Cre+ β2fl/+	Pax6α-Cre+ β2fl/−	β2−/− + cAMP
Retinal wave frequency (waves/min)	1.00 +/−0.06 (n = 27)	0.34 +/− 0.05 (n=56) ***	1.23 +/−0.08 (n=8)		1.14 +/−0.07 (n=3)	1.01 +/− 0.15 (n=8)	1.15 +/− 0.12 (n=5) $$$
SC wave frequency (waves/min)	1.21 +/−0.06 (n = 16)	3.28 +/− 0.36 (n=22)***, ###	1.32 +/−0.06 (n=2)	1.12 +/−0.06 (n=5)	0.92 +/−0.11 (n=7)	0.74 +/−0.12 (n=7)	4.78 +/−1.18 (n=4) ***
Retinal Wave Speed (μm/s)	83.2 +/−7.4 (n=16)	180 +/−10.9 (n=31) ***	78.7 +/−8.0 (n=8)		86.2 +/−10.6 (n=3)	78.6 +/−7.6 (n=8)	207.1 +/−34.6 (n=5) ***
SC Wave Speed (μm/s)	67.3 +/−7.6 (n=16)	259 +/−13 (n=22) ***, ###	66.1+/−2.2 (n=2)	59.8 +/−5.9 (n=5)	67.5 +/−11.6 (n=7)	96.7 +/−11.3 (n=7)	254.8 +/−34.1 (n=4) ***
Retinal Wave Duration (s)	20.4 +/−1.7 (n=16)	9.6 +/−0.5 (n=31) ***	21.3 +/−1.6 (n=8)		23.3 +/−1.5 (n=3)	11.4 +/−1.3 (n=8) *	11.2 +/−0.9 (n=5) ***
SC Wave Duration (s)	27.7 +/− 1.6 (n=16) ###	4.9 +/−0.4 (n=22) ***, ##	21.7 +/−0.1 (n=2)	26.8 +/−2.7 (n=5)	22.6 +/−3.4 (n=7)	12.1 +/−0.9 (n=7) **	5.8 +/−0.4 (n=4)***
Retinal Wave Size (mm2)	0.62 +/− 0.08 (n=16)	1.08 +/− 0.07 (n=31) ***	0.99 +/− 0.08 (n=8)		0.84 +/− 0.03 (n=3)	0.38 +/− 0.04 (n=8) *	1.23 +/− 0.1 (n=5)**
SC Wave Size (mm2)	0.93 +/−0.05 (n=16) #	0.45+/−0.02 (n=22) ***, ###	0.96+/−0.07 (n=2)	1.06 +/−0.09 (n=5)	0.64 +/−0.12 (n=7)	0.22 +/−0.07 (n=7) **	0.77 +/−0.11 (n=4)***, $$
Wave Direction Bias Index	0.31 +/−0.04 (n=16)	0.61 +/−0.05 (n=29) ***	0.35 +/−0.05 (n=8)		0.35 +/−0.06 (n=13)	0.26 +/−0.02 (n=11)	0.50 +/−0.13 (n=5)
Retinal ROI Wave Frequency (waves/min)	0.69 +/−0.09 (n=16)	0.39 +/−0.05 (n=32) **	0.94 +/−0.1 (n=8)
Retinal ROI Inter-wave Interval (s)	105 +/−13.4 (n=16)	169 +/−17.1 (n=32) *	68.3 +/−6.9 (n=8)
Retinal ROI Wave Duration (s)	0.99 +/−0.04 (n=16)	1.02 +/−0.02 (n=32)	1.14 +/−0.07 (n=8) *
Retinal ROI Wave Amplitude (ΔF/F)	0.032 +/− 0.002 (n=16)	0.021 +/−0.001 (n=32) ***	0.024 +/− 0.001(n=8) *
Retinal wave frequency (waves/min) P3	1.00 +/−0.14 (n = 6)	0.27 +/− 0.02 (n=4) **	1.48 +/−0.13 (n=2)
Retinal wave frequency (waves/min) P4	0.84 +/−0.14 (n = 6)	0.19 +/− 0.05 (n=11) **	1.12 +/−0.02 (n=2)
Retinal wave frequency (waves/min) P5	0.99 +/−0.12 (n = 6)	0.18 +/− 0.04 (n=15) ***
Retinal wave frequency (waves/min) P6	1.09 +/−0.09 (n = 5)	0.33 +/− 0.07 (n=17) ***	1.30 +/−0.10 (n=2)
Retinal wave frequency (waves/min) P7	1.14 +/−0.07 (n = 4)	0.81 +/− 0.17 (n=9)	1.00 +/−0.10 (n=2)

^*p<0.05,

^**p<0.01,

^***p<0.001. Significant difference between condition and equivalent WT property.

^#p<0.05,

^##p<0.01,

^###p<0.001. Significant difference between presynaptic and postsynaptic within condition/genotype.

^$p<0.05,

^$$p<0.01,

^$$$p<0.001. Significant difference between β2^−/− before and after CPT-cAMP application.

Measurements are aggregated by hemisphere and reported as means (+/− standard error of the mean). Statistical significance was determined by using one or two-way analysis of variance (ANOVA) along with Tukey’s honest significant difference (HSD) post-hoc test for reported p-values. (Relates to Figures 1–7, S6).

Activity in SC Neurons Doesn’t Match Retinal Activity in β2^−/− Mice

Spontaneous retinal waves are thought to guide the refinement of developing neural circuits through a Hebb-based mechanism that relies on the correlation between presynaptic retinal ganglion cell activity and postsynaptic neuronal activity. In WT mice, retinal waves produce corresponding waves of activity in postsynaptic neurons (Ackman et al. 2012). To determine whether these correlations persist in β2^−/− mice in light of the disrupted retinal wave activity, we examined spontaneous activity in SC neurons using exogenous application of a calcium fluorophore (OGB1AM) and/or genetically encoded calcium indicators expressed in SC neurons but not the retina (En1-Cre X floxed-GCaMP3; Fig. 3). In WT mice, the spatial and temporal properties of spontaneous activity was very similar in recordings of presynaptic retinal afferents and postsynaptic neurons in the SC (Fig. 3B, G, Movie S2), consistent with a dominant role for retinal waves in driving postsynaptic activity in the SC before eye opening (Ackman et al., 2012). Remarkably, in β2^−/− mice there was a dramatic difference in presynaptic activity recorded in retinal afferents and postsynaptic activity observed in SC neurons. In the SC of β2^−/− mice, in addition to larger and faster, but infrequent waves, as found in retinal afferents of β2^−/− mice (Fig. 2, Fig. 3B), we observed a new, distinct form of spontaneous activity that was frequent, very rapid, and largely confined to the midline between the two hemispheres (Movie S2, Fig. 3C, Fig. S3). This unexpected ‘extra’ pattern of spontaneous activity accounted for the vast majority of the postsynaptic activity in the SC of β2^−/− mice and dramatically increased the overall frequency of spontaneous activity to levels well above that observed even in WT SC (Fig. 3D, E). As with activity recorded from retinal ganglion cell afferents, most spontaneous activity parameters recorded from SC neurons were significantly different in β2^−/− mice relative to WT mice. This included wave size, duration and speed (Fig. 3F, Fig. S3, Table 1), with all of the parameters consistent with the appearance in β2^−/− SC neurons of very small, transient, and fast midline waves, which were not typically observed in WT mice (Fig. S3). For example, in WT mice, there was a clear correspondence between wave size and wave duration, with smaller waves being of shorter duration and larger waves having longer duration in both presynaptic RGC afferents and postsynaptic SC neurons (Fig. 3F). In sharp contrast, in β2^−/− mice a significant correlation between wave size and duration was lacking for RGC activity (Fig. 3G), while in β2^−/− SC neurons the appearance of an ‘extra’ class of very small and brief waves restored the correspondence between wave size and duration, and completely disrupted the match between presynaptic RGC activity and postsynaptic SC waves found in WT recordings.

Figure 3. Spontaneous activity in the SC is largely independent of retinal activity in β2^−/− mice.

A. Neuronal activity is measured in SC neurons by crossing En1-Cre to floxed GCaMP3 mice or loading OGB1AM-dye into the SC. B. Example SC wave activity montages in WT (top –P7) and β2^−/− (bottom – P3) mice. Activity in WT mice (top) has clear propagating wave front (white arrows). SC Activity in β2^−/− mice displays two characteristic patterns, infrequent “large” waves (bottom panel, one wave in each hemisphere) that are similar to retinal activity in β2^−/− mice (Fig. 1) and (C.) more frequent, highly stereotyped and bilaterally correlated “small” waves that propagate rapidly in the anterior and medial SC (P3–4). Scale = 200 μm. D. Example raster plots of SC activity in WT (top –P3) and β2^−/− mice (bottom – P3). E. Summary quantification of retinal (Ret) and SC wave frequency in WT and β2^−/− mice. Waves measured in SC neurons in β2^−/− mice are much more frequent than retinal waves in β2^−/− mice or waves in WT mice. F. There is a strong and similar correspondence between wave size and duration in both RGCs and SC neurons in WT mice (WT retinal wave size vs. duration r=0.71, n = 390 waves; WT SC wave size vs. duration r=0.63, n = 298 waves, p<0.001). G. Retinal wave size and duration are unrelated in β2^−/− mice (r=0.23, n = 338 waves), but SC wave size and duration remain correlated (r=0.62, n = 878 waves, p<0.001), leaving spontaneous activity totally mismatched between RGCs and SC neurons in β2^−/− mice. See also Figure S3.

In summary, these data indicate that coordinated spontaneous retinal activity is dramatically reduced in frequency in β2^−/− mice, and that the properties of the remaining spontaneous activity are fundamentally altered. In the SC, an apparently new form of spontaneous activity appears in β2^−/− mice that is markedly different in form from both WT waves and the altered presynaptic activity observed in β2^−/− RGCs, potentially due to homeostatic pressures in the absence of coordinated spontaneous drive from the retina.

β2-nAChRs in the SC are Not Required for Normal Wave Activity

β2-nAChRs are expressed throughout the developing nervous system and are known to contribute to glutamatergic synapse development in the CNS (Lozada et al., 2012). The very unusual properties of spontaneous retinal waves and the mismatched activity in developing SC neurons in whole-animal β2^−/− mice in vivo creates some ambiguity about the role of retinal waves and the specific contribution of β2-nAChRs in the retina or SC to visual map development. To resolve this ambiguity, we generated mice with loxP sites flanking exon 5 of Chrnb2 (Fig. 4A, B), the largest exon of the β2-nAChR gene encoding half of the N-terminal ligand binding domain and 3 of 4 transmembrane domains. We confirmed that germ line deletion of exon 5 produced functional null receptors in the retina (Fig. 4C) and recapitulated visual system anatomical phenotypes previously reported in whole-animal β2^−/− mice (Fig. S4). We then crossed floxed β2-nAChR mice to Cre recombinase lines that are specific to either midbrain postsynaptic (En1-Cre⁺;β2^fl/− or SCβ2-cKO mice, Fig. 4D) or retinal presynaptic (Pax6α-Cre⁺;β2^fl/− or Retβ2-cKO mice, Fig. 5, 6) cell populations. Crossing midbrain-specific En1-Cre mice to floxed β2-nAChR mice to generate SCβ2-cKO mice completely eliminated β2-nAChR mRNA expression in the SC, but retinal mRNA levels were unchanged (Fig. 4E). Wave properties in SC neurons in vivo in the SCβ2-cKO mice were indistinguishable from WT mice (Fig. 4F), with normal frequency, duration, and speed, but were significantly different from β2 whole animal knockout (β2^−/−) waves (Fig. S5). Anatomical analysis of retinocollicular projections using focal retinal injections of DiI in the retina (to assess retinotopic map formation in the SC) or whole-eye injections of CTB-Alexa 488 and 555 (to study eye-specific segregation in the SC) at P6 revealed that retinotopic mapping and eye-specific segregation in SCβ2-cKO mice were indistinguishable from control mice at postnatal day 8 (Fig. 4G, H, Fig. S5). These experiments demonstrate that β2-nAChRs in SC neurons have no effect on spontaneous activity patterns or early visual circuit development.

Figure 4. Conditional deletion of β2-nAChRs from the SC has no effect on wave activity, retinotopy, or eye-specific segregation.

A. Design of floxed β2-nAChR mice (also see Experimental Procedures and Figure S4). B. Southern blot showing WT, β2^−/− and floxed-β2 (β2^fl/fl) bands (WT= 8.0 kB; β2^−/− = 5.8 kB; β2^fl/fl = 4.7 kB). C. Knockout of β2-nAChR subunit eliminates response to a nicotinic acetylcholine receptor agonist (DMPP) in starburst amacrine cells in the retina of a P5 mouse, 1mM, 300 ms puff. D. Expression of En1-Cre, visualized through lacZ or GFP reporter mice, at embryonic and early postnatal ages. En1-Cre was used to delete β2-nAChR subunits from the midbrain but not the retina of β2^fl/− mice. E. Reverse transcription PCR reveals that mRNA for β2-nAChR is absent from the SC of En1-Cre⁺;β2^fl/− (SCβ2-cKO) mice, but is spared in the retina at P6–7. F. Example montage of SC wave activity in SCβ2-cKO mice (P5 example). Waves in SCβ2-cKO mice are indistinguishable from waves in WT mice in both frequency and spatiotemporal properties (see Fig. S4 and text). G. Retinotopy is unaltered in SCβ2-cKO mice. Example target zones (yellow spots) in the SC (white dotted outlines) following focal injections of DiI into the ventral-temporal (left-top) or dorsal (left-bottom) retina. Summary quantification of target zone size shown on the right for ventral-temporal retinal injections (right - top) and dorsal injections (right - bottom). Scale = 500 μm. H. Eye-specific segregation is unaltered in SCβ2-cKO mice. Example ipsi (top – grey scale) and contra (green) with ipsi (red) projections together (bottom) in sagittal sections through the SC. Summary quantification of eye segregation (right), measured as the fraction of SGS layer (dotted outline) with ipsilateral label. Scale = 500 μm. See also Figures S4, S5.

Figure 5. Conditional deletion of β2-nAChRs from the retina eliminates waves in the retina and SC.

A. Region-specific expression of Pax6α-Cre in the retina (GFP reporter – top panel; cartoon – middle panel). Cre is expressed in the peripheral two-thirds of the retina in Pax6α-Cre⁺ mice, and GFP fluorescence in the SC (bottom panel) shows corresponding Pax6α-Cre⁺ (anterior and posterior) and Pax6α-Cre⁻ (medial-lateral strip) RGC axon projection regions in the SC. B. Example retinal wave montage from Pax6α-Cre⁺;β2-nAChR^fl/+ control (heterozygous) mouse (P7). C. Map of average activity in this control example reveals that waves extend throughout the SC. D. Example retinal wave montage from a Retβ2-cKO (Pax6α-Cre⁺;β2-nAChR^fl/−) mouse (P5). E. Map of average activity shows that retinal waves in Retβ2-cKO mice are confined to regions of the retina lacking Cre expression (Pax6α-Cre⁻). Scale = 200μm. Retinal wave activity in Retβ2-cKO mice shows a sharp border at the posterior edge of the Cre⁻ region, which is absent in control (Pax6α-Cre⁺;β2-nAChR^fl/+) littermate mice. Note that the cortex progressively covers the anterior-lateral aspect of the SC during the first week after birth. F. Regional analysis of wave frequency in Retβ2-cKO and littermate control mice demonstrates that wave activity is restricted to the Cre⁻ region in Retβ2-cKO mice, but is spread evenly in littermate control mice. G. Wave speed in Retβ2-cKO mice is similar to WT and littermate control mice and much slower than β2^−/− mice. H. Wave size is smaller in Retβ2-cKO (Pax6α-Cre⁺;β2-nAChR^fl/−) relative to WT, β2^−/− and littermate control mice. I. Wave direction bias index in Retβ2-cKO (Pax6α-Cre⁺;β2-nAChR^fl/−) mice is indistinguishable from WT and littermate control mice (Pax6α-Cre⁺;β2-nAChR^fl/+) but β2^−/− waves are more directionally biased than any other group. See also Table 1, Figure S6.

Figure 6. Conditional deletion of β2-nAChRs from the retina disrupts circuit development in a region-specific manner.

A. Cartoon illustrating region-specific retinotopic mapping defects in WT, β2^−/−, and Retβ2-cKO mice. B. Retβ2-cKO retinotopy is specifically disrupted in regions of the SC (ventral-nasal –left panels; ventral-temporal – right panels) that receive projections from RGCs that lack β2-nAChR subunit expression and waves at P8. Retinotopy of RGCs from the dorsal retina in Retβ2-cKO mice, which maintain expression of β2-nAChRs and retinal waves, is unperturbed relative to littermate controls (middle panels). Bottom row of histograms shows summary quantification of retinotopic defects in Retβ2-cKO relative to littermate control mice for ventral-temporal (left), dorsal (middle) and ventral-nasal (right) retinocollicular projections. C. Eye-specific segregation is disrupted in the SC of Retβ2-cKO mice relative to littermate control mice at P8. Grey scale panels show ipsilateral projections at two magnifications, and color panels show ipsi (red) and contra (green) projections (scale = 500 μm). Summary quantification of eye segregation in the SC shown in histograms on right. D. Eye-specific segregation is also disrupted in in the dLGN of Retβ2-cKO mice relative to littermate control mice at P8. Ipsilateral projection shown in red, contralateral in green. Overlap of the two eyes projections in the dLGN is shown in white (scale = 500 μm). Histograms on right show summary quantification of eye segregation in the dLGN. In all histograms, Retβ2-cKO are shown in red, littermate controls shown in black and green.

β2-nAChRs in the Retina are Required for Normal Wave Activity

To examine the role of retinal β2-nAChRs on retinal wave propagation and visual circuit development, we generated retina-specific β2-nAChRs mutants by crossing Pax6α-Cre mice with floxed-β2 mice (Pax6α-Cre⁺;β2^fl/− or Retβ2-cKO mice). Cre recombinase expression in Pax6α-Cre mice is high in the peripheral retina, but weak in the central retina along a dorsal-ventral strip (Marquardt et al. 2001; Fig. 5A). This delimits the retina into three broad subregions, with RGCs in the temporal and nasal retina, which project to the rostral and caudal SC, expressing Cre recombinase, while RGCs in the central retina, which project to a medial-lateral strip in the middle of the SC, largely spared of Cre recombinase activity and serve as an internal control (Fig. 5A). The clear borders between Cre-expressing sub-regions of the retina permit the identification of regions of the SC that lack β2-nAChR expressing RGCs (rostral and caudal SC) and control (mid SC) RGC axon-projecting regions that maintain β2-nAChR expression (Fig. 5A bottom). Remarkably, wave activity in Retβ2-cKO mice observed in both RGC axons and SC neurons was limited to the medial-lateral strip in the SC containing RGCs with maintained β2-nAChR expression (Fig. 5D, E), and waves neither nucleated nor propagated into the knockout regions of the retina (Movie S3). A quantitative analysis of wave frequency in the rostral and caudal SC shows almost no wave activity in these regions in Retβ2-cKO mice, while wave activity in the medial SC was unchanged relative to control mice in Retβ2-cKO mice (Fig. 5F, Table 1). Interestingly, the waves that persisted in Retβ2-cKO mice were spatiotemporally similar to waves observed in control littermate (Pax6α-Cre⁺;β2^fl/+) and WT mice (Fig. 5G–I, Fig S6; Table 1), with a clear propagating wave front and wave speeds that were comparable to WT and control littermate waves (Fig. 5G, Table 1). Wave size was significantly smaller in Retβ2-cKO than littermate control, β2^−/− and WT mice (Fig. 5H, Table 1), and wave duration was much shorter than in WT and littermate control mice (Fig. S6A, Table 1), presumably because wave propagation was truncated upon entering the null region of the retina in Retβ2-cKO mice. Wave direction bias in Retβ2-cKO was also indistinguishable from WT and littermate control mice, and significantly less than the bias in β2^−/− mice (Fig. 5I, Table 1). These data demonstrate that regions of the retina that lack expression of β2-nAChRs also lack wave activity, while normal propagating retinal waves persist, in the same retinas, where β2-nAChR expression persists.

We next used anatomical techniques to examine visual circuit development in the Retβ2-cKO mice. To examine retinotopic map development, we made targeted focal injections of DiI around the periphery of the retina (Fig. 6A). In Retβ2-cKO mice, projections to the SC and dLGN from dorsal RGCs, which lack Cre recombination and maintain normal retinal wave activity, were indistinguishable from littermate control mice at P8 (Fig. 6B, Fig. S6C). In stark contrast, RGC projections from ventral-temporal and ventral-nasal retina in Retβ2-cKO mice, with robust Cre recombination and no retinal wave activity, were dramatically disrupted. Thus, regions of the retina in Retβ2-cKO with normal retinal waves maintained normal retinotopic projections, while retinotopic refinement in regions of the retina without retinal waves was profoundly disrupted (Fig. 6A, B). These experiments provide a strong causal link between retinal wave activity and the development of normal retinotopic projections.

We next examined the emergence of eye specific segregation in the SC and dLGN of Retβ2-cKO and littermate control mice. As with the development of retinotopy, eye-specific segregation in the SC was disrupted in Retβ2-cKO mice in comparison to all littermate control mice at P8 (Fig. 6C). Moreover, eye-specific segregation in the dLGN, like in the SC, was significantly worse in Retβ2-cKO mice than in littermate controls (Fig. 6D). These experiments provide definitive support of a necessary role for retinal waves in the emergence of both retinotopy and eye-specific segregation in the developing SC and dLGN.

Rescuing Retinal Waves Restores Eye-Specific Segregation in β2^−/− Mice

Our data suggest that the loss of spontaneous retinal activity leads to axon refinement and neural circuit defects in whole animal (β2^−/−) and retina-specific (Retβ2-cKO) β2-nAChR conditional knockout mice. To directly test this hypothesis, we attempted to ‘rescue’ spontaneous retinal waves in β2^−/− mice and examine the effects on in vivo activity patterns and RGC axon refinement. CPT-cAMP increases the frequency of retinal waves in neonatal ferret in vitro (Stellwagen et al., 1999), and improves eye-specific segregation in mice in vivo (Xu, Furman et al., 2011). Application of CPT-cAMP to the retina of β2^−/− mice dramatically increased the frequency of in vivo spontaneous retinal activity (Movie S4, Fig. 7A–C). Interestingly, while the frequency of spontaneous retinal activity in β2^−/− mice following treatment with CPT-cAMP increased to levels that were indistinguishable from control mice (p = 0.951; Fig. 7C and Table 1), the waves remained highly correlated between the eyes (Fig. S7) and the spatiotemporal properties of the induced retinal activity remained disrupted. In particular, retinal wave speed, duration, and size (Fig. 7D, E, Fig. S7 and Table 1) were similar with and without CPT-cAMP treatment in β2^−/− mice, but were significantly different from control mice. Chronic daily treatment of β2^−/− mice with CPT-cAMP in both eyes (vitreal application of 2mM CPT-cAMP from P2-P6) significantly improved eye-specific segregation in both the SC (Fig. 7F) and dLGN (Fig. 7G) in comparison to saline treated β2^−/− mice at P8. However, retinotopic refinement was not improved by CPT-cAMP treatment in either β2^−/− mice or WT mice (Fig. 7H), suggesting that CPT-cAMP’s effect is not a generalized consequence of actions on neurite outgrowth and branching (Nicol et al., 2007), but are limited to eye-specific segregation. These data provide strong evidence that a threshold level of coordinated spontaneous retinal activity is necessary for the emergence of eye-specific segregation, but appropriate spatiotemporal patterns of retinal waves are required for retinotopic refinement.

Figure 7. Eye-specific segregation is partially rescued after increasing frequency of retinal spontaneous activity in β2^−/− mice.

A. Experimental schematic. Activity recordings are made in the SC of whole-animal β2^−/− mice before and after double-eye injection of CPT-cAMP (2 mM, ~500nL). B. Example wave activity raster plot from 5 minute recording before (top) and after (bottom) double eye CPT-cAMP injection. Wave frequency is increased in the SC contralateral to the eye that receives CPT-cAMP. C. Summary of effects of CPT-cAMP on wave frequency in β2^−/− mice. Application of CPT-cAMP to β2^−/− mice restores wave frequency to levels comparable to WT mice, which is much higher than β2^−/− mice. D. Wave speed in β2^−/− CPT-cAMP treated retinas is no different from wave speed in β2^−/− mice without CPT-cAMP. E. Wave duration in β2^−/− CPT-cAMP treated retinas is no different from wave duration in β2^−/− mice without CPT-cAMP. F. Eye-specific segregation is partially rescued at P8 in the SC and dLGN (G) of β2^−/− mice after 5 days of binocular CPT-cAMP treatment (from P2-P6). H. In contrast, retinotopic mapping in the SC is unaffected in both WT and β2^−/− mice after chronic CPT-cAMP treatment at P8 (scale = 500 μm). See also Figure S7.

DISCUSSION

Molecular and activity-dependent factors act together during development to guide the formation of refined neural circuits. Here, we describe experiments that examined the role of retinal β2-nAChRs and spontaneous retinal waves in visual circuit development. We focused our experiments on characterizing the properties of spontaneous activity in the intact, unanesthetized visual system in vivo, because previous in vitro experiments produced conflicting results about the role of β2-nAChRs in retinal wave generation (Bansal et al., 2000; McLaughlin et al., 2003; Sun, Warland et al., 2008; Stafford et al., 2009), and our own experiments show that the spatiotemporal properties of retinal waves differ to some extent in vivo and in vitro (Ackman et al., 2012). We observed that the frequency of activity and the dynamics of retinal waves were dramatically altered in β2^−/− (whole animal knockout) mice. Moreover, unlike WT mice, spontaneous activity in neurons in the developing superior colliculus no longer resembled retinal activity in β2^−/− mice, suggesting that neurons in the SC develop an intrinsic pattern of activity in the absence of adequate retinal drive. We also used conditional genetic deletion of the β2-subunit to demonstrate that midbrain β2-nAChRs are unimportant for retinal waves and anatomical circuit refinement, but that retinal knockout of β2-nAChRs eliminated waves in both RGCs and SC neurons and greatly disrupted visual circuits in a region-specific manner. We finally showed that boosting retinal wave frequency in β2^−/− mice using a pharmacological approach was sufficient to partially rescue eye-specific segregation in both the dLGN and SC, but had no effect on retinotopy. Overall, our experiments underscore the strong causal link between retinal waves and normal visual circuit development, specifically implicate retinal β2-nAChRs in this process, and distinguish between a role for overall retinal wave frequency and dynamics in distinct aspects of visual circuit refinement in the developing dLGN and SC of mice.

Spontaneous Activity in vivo vs. in vitro

Retinal waves are a network property governed by cellular physiology and circuit dynamics under broad neuromodulatory influence. In this context, it is perhaps not surprising that some aspects of wave dynamics differ in vitro and in vivo. These differences may be exacerbated in mice (like the β2^−/− mice) in which the main neurotransmitter system responsible for the predominant form of network activity is mutated. Under in vitro conditions, reports variously suggest that retinal waves in β2^−/− mice are either absent (Bansal et al., 2000; McLaughlin et al., 2003), present at a much higher frequency (Sun, Warland et al., 2008), or present with a modestly reduced frequency (Stafford et al., 2009). Our data clearly demonstrate that waves are much less frequent in vivo in β2^−/− mice than WT mice, which is consistent with Stafford et al. (2009). However, when observed in vitro, waves in β2^−/− mice lack directional bias, but are of similar duration as waves in WT mice (Stafford et al., 2009). Neither of these network properties was observed in vivo, as retinal waves in β2^−/− mice are much shorter in duration (Fig. 2), and show an even more pronounced directional bias than WT mice (Fig. 2). Other β2^−/− wave properties observed in vitro are similar to what we observed in vivo, including wave dependence on Gap junctions (Sun, Warland et al., 2008) and stronger correlations over a larger area of the retina (Stafford et al., 2009; Sun, Warland et al., 2008). We also observed a remarkable increase in wave correlations between the eyes in β2^−/− mice, which obviously could not be predicted from in vitro experiments. Finally, we note that the somewhat contradictory reports on the role of retinal waves in visual circuit development (e.g. Penn et al., 1998; Huberman et al., 2003; Sun, Warland et al., 2008; Chalupa, 2009; Feller, 2009) rely in part on extrapolating in vitro analysis of retinal network dynamics after pharmacologic or genetic manipulations to the in vivo condition. Our data suggest that this inference is potentially problematic, and highlights the importance of examining network properties of spontaneous activity in an intact, in vivo condition.

Examining Retinal Waves in RGC Axon Arbors in the SC

We examined the spatiotemporal dynamics of retinal waves by measuring activity in retinal axon arbors in the SC. Differences in RGC axon arbors in β2^−/− mice relative to WT mice (Dhande et al. 2011) are unlikely to contribute significantly to altered retinal wave dynamics observed in β2^−/− mice for a number of reasons. First, axon arbors in β2^−/− mice are expanded relative to WT mice at P8, but are indistinguishable from WT mice at P4 (Dhande et al. 2011), whereas we observed that retinal waves are altered in β2^−/− mice throughout the first week after birth. Similarly, RGC axon arbors in WT mice refine significantly throughout the first postnatal week (Dhande et al. 2011), but wave dynamics are no different in WT mice at P3 than at P7. Second, enlarged RGC axon arbors in β2^−/− mice should have no effect on retinal wave duration, whereas we observed spontaneous activity that was dramatically shorter in duration in β2^−/− mice relative to control mice. Third, the dynamics of retinal waves (speed, size, directionality, etc.) in WT mice measured in RGC axon terminals in the SC are very similar to the wave properties measured in SC neurons, suggesting that RGC axon arbors do not significantly distort retinal waves. Finally, and perhaps most tellingly, total wave size (area) is about 30 times larger than the area of a given RGC axon arbor in both β2^−/− and WT mice (Dhande et al., 2011). Thus, wave size is governed much more by the area of wave propagation in the retina than it is by the size of a given RGC axon arbor in the SC. Similarly, the speed of a propagating wave in the retina, as measured in the SC, is not significantly altered by changes in the size of the RGC axon arbor in the SC, since the axon arbor is so much smaller (and the Ca²⁺ transients associated with the wave passing through much briefer) than the total area consumed by a wave. Perhaps the only wave parameter that we expect may be altered by having a larger axon arbor is the size of the wave front as it propagates across the SC, and in fact it is often difficult to even distinguish a clear wave front in β2^−/− mice in comparison to WT mice (Fig. 1B) Overall, these various lines of evidence suggest that the altered wave dynamics we observed in RGC axon terminals in the SC of β2^−/− mice reflect genuine differences in spatiotemporal patterns of activity in the retina.

Why are Visual Maps Disrupted in β2^−/− Mice?

Whole animal β2^−/− mice show a remarkably strong anatomical and functional phenotype, and have become the most commonly used ‘retinal wave mutant’ for the study of activity-dependent visual circuit refinement (Kirkby, Sack et al., 2013; Cang and Feldheim, 2013; Ackman and Crair, 2014). Despite the severity of their visual circuit disruption and their popularity as a model system, the specific cause of the various β2^−/− mutant phenotypes was unclear. Our experiments specifically localize the cause of visual circuit and functional deficits in β2^−/− mice to disruption in the retina and retinal activity, but a priori we still imagine four distinct factors that may individually or in combination explain the visual circuit phenotypes observed in β2^−/− mice. First, the dramatically reduced wave frequency, especially early in the first week when RGC axon elaboration is at its peak (Dhande et al., 2011), may disrupt circuit refinement. Second, the dissociation between presynaptic (retinal) waves and postsynaptic (SC neuron) activity may interfere with Hebbian (‘fire together, wire together’) mechanisms for circuit refinement. Third, the infrequent coordinated spontaneous ‘wave’ activity that persists in the retina of β2^−/− mice has severely disrupted spatiotemporal properties, which may negatively influence normal circuit refinement. Finally, the few retinal waves that do persist in β2^−/− mice are highly correlated between the two retinas, which may also disrupt circuit refinement, particularly with respect to eye-specific projections. Our experiments using chronic CPT-cAMP application to boost retinal wave frequency in β2^−/− mice specifically implicate retinal wave frequency for developmental disruptions in eye-specific segregation (Fig. 7). However, CPT-cAMP application did not completely correct eye-specific segregation in β2^−/− mice, and retinotopic circuits remained totally disrupted, which suggests that both a minimum level of spontaneous coordinated wave activity as well as specific patterns of spontaneous activity are together necessary for normal visual circuit refinement (Zhang et al., 2011; Xu, Furman, et al., 2011; Godfrey et al., 2009; Butts et al., 2007).

Changes in Spontaneous Activity in the SC of β2^−/− Mice

Spontaneous retinal waves in β2^−/− mice are GAP junction mediated, significantly reduced in frequency, and gradually recover to near WT levels by the end of the first postnatal week after birth. This suggests that compensatory mechanisms encourage the persistence of coordinated activity in the retina, despite their dramatically altered form relative to WT mice. There is normally a strong correspondence between spontaneous (presynaptic) retinal wave activity and (postsynaptic) SC neuron activity, while in β2^−/− mice there is a marked disconnect between retinal activity and collicular neuron activity. The patterned spontaneous activity in the SC of β2^−/− mice dissociated from retinal activity is strikingly similar to the pronounced rhythmic spiking activity observed in the dLGN of young enucleated ferrets (Weliky and Katz, 1999), which may explain the preservation of structured geniculocortical and corticocortical connections in enucleated ferrets (Ruthazer and Stryker, 1996; Crowley and Katz, 1999). The persistent correlated spontaneous activity in the dLGN of enucleated ferrets relies on feedback projections from the visual cortex (Weliky and Katz, 1999). Similarly, descending cortical input to the SC may be the source of some of the collicular activity in β2^−/− mice. However, these projections normally don’t emerge until P6-P8 (Triplett et al., 2009), while the unusual SC activity is already present at P3 in β2^−/− mice. Dynamic reorganization on a network level may be a common phenomenon in the early neonatal brain (Stacy et al., 2005; Kirkby and Feller, 2013), and could facilitate large scale compensatory rewiring in response to disruptions in neural activity resulting from a variety of neurodevelopmental insults and disorders (Ebert and Greenberg, 2013).

Region Specific Effects

Regions of the retina with persistent β2-nAChR expression (in Retβ2-cKO mice) maintain relatively normal wave frequency and spatiotemporal properties both in the retina and target regions of the SC (Fig. 5), whereas β2-nAChR ‘null’ regions of the retina are nearly devoid of any spontaneous activity (but see Stacy et al., 2005). Anatomical projections in Retβ2-cKO mice are indistinguishable from WT mice only in the regions of the retina that maintain normal waves, strengthening the causal link between retinal waves and normal visual circuit development. It is notable that waves in the null part of the retina are nearly absent; whereas in β2^−/− mice waves are significantly reduced in frequency, but persist, albeit with quite abnormal spatiotemporal dynamics. This suggests that compensatory mechanisms keep track of activity across the entire (or large swaths) of) the retina, and maintain Gap junction mediated waves in β2^−/− mice (Stacy et al., 2005; Kirkby and Feller, 2013), but are silent in β2-nAChR ‘null’ regions of Retβ2-cKO mice.

Bilateral Correlations in Retinal Activity in β2^−/− Mice

We previously reported that WT mice exhibited unanticipated, but weak, bilateral correlations in retinal wave activity (Ackman et al., 2012). These bilateral correlations were dramatically enhanced in β2^−/− mice relative to control mice (Fig. 2). Similarly, retinal waves in WT mice exhibit a marked directional bias, which is also enhanced in β2^−/− mice. The directional bias may be related to the tendency for waves to nucleate in the ventral-temporal retina, which could be enhanced in β2^−/− mice. The origin of the bilaterally correlated activity also remains uncertain, but we note that bilateral activity sometimes (though not always) appeared to begin in one retina and then propagate with a small temporal lag to the contralateral retina (e.g. Fig. 3B). This suggests that direct synaptic communication between the retinas in neonatal mice may be the origin of some correlated bilateral activity (Bunt and Lund, 1981; Ackman et al., 2012). β2^−/− mice may have more correlated binocular activity through enhanced retino-retinal connections or by virtue of the larger waves observed in β2^−/− mice more frequently engaging retino-retinal communication. Bilateral correlations in spontaneous activity are unlikely to be the (sole) origin of disrupted visual circuit development in β2^−/− mice because CPT-cAMP application improved eye-specific segregation by enhancing retinal wave frequency, but spontaneous activity remained correlated between the two eyes even after this pharmacological manipulation (Fig. S7). Finally, we note that the partial recovery of eye-specific segregation but not retinotopy in the SC and dLGN of β2^−/− mice treated with CPT-cAMP mice further suggests that the presence, frequency and spatiotemporal pattern of retinal waves are all important for normal visual circuit refinement, and that different aspects of circuit development are dependent on different features of spontaneous activity (Zhang et al., 2011; Xu, Furman et al., 2011).

Conclusions

In summary, these experiments establish a firm causal link between nicotinic acetylcholine receptors, retinal waves and visual circuit development, and demonstrate that spontaneous retinal waves are necessary and instructive for circuit refinement in the developing nervous system. They also validate the use of β2^−/− mice as a model for disrupted spontaneous activity during development, though multiple altered features of retinal waves in these mice may contribute to circuit refinement defects. The persistence of patterned spontaneous activity in retinorecipient regions of the brain following the withdrawal of retinal input, as was observed in β2^−/− mice and enucleated ferrets (Weliky and Katz, 1999), suggests that homeostatic pressures work to preserve network activity and adapt to the loss of sensory drive, thus generating structured neural circuits that reflect the altered neonatal activity patterns. These results also suggest that normal variations in spontaneous activity can lead to the emergence of a range of functional circuits, and that disrupted patterns of spontaneous activity early in development may play an important role in the etiology of circuit-based neurodevelopmental disorders.

EXPERIMENTAL PROCEDURES

Animals

We created a line of floxed β2-nAChR mice in order to conditionally and specifically delete β2-nAChR2. β2-floxed mice were made following published strategies (Xu et. al. 2009; Xu, Cohen et al. 2011) using DNA from a BAC clone containing large fragments of the Chrnb2 genomic DNA from the 129SvEv mouse genome. LoxP sites were inserted surrounding exon 5, the largest coding exon (970bp) of the gene, and conditional deletion after crossing to Cre recombinase mice was confirmed using standard reverse transcriptase PCR (see Supplemental Experimental Procedures, Figs. 4–6 and Fig. S4). Littermate controls blind to the genotype were used everywhere, except in supplementary figure 5, where, due to the difficulty of the experiments and breeding scheme, pre-genotyping was used to focus on En-Cre⁺;β2^f/− mice.

In vivo Calcium Imaging

Spontaneous activity was recorded from neonatal mice as has been described in detail previously (Ackman, 2012; also see Supplemental Experimental Procedures). Calcium dyes were introduced either through injection of exogenous calcium dye (Calcium Green-1 Dextran, Invitrogen #C-6765, or Oregon Green 488 Bapta-1 AM, Invitrogen #O-6807) into the retina/SC, or induction of genetically-encoded GCaMP through either viral injection in the retina (GCaMP6, Penn Vector Core, AV-1-PV2833) or crossing of a floxed GCaMP3 mouse (Ai38, Jackson Laboratory #014538) to either retina (Rx-Cre, Dhande et al., 2012) or midbrain (En1-Cre, Dhande et al., 2012) - specific Cre lines. After retinal injections, calcium dye was allowed to fill retinal ganglion cell axons for at least two (Calcium Green-1) or four (viral GCaMP6) days. Imaging of activity was performed on unanesthetized, head-fixed mice after a 1–2 hour recovery from isoflurane anesthesia and the surgical procedure to expose the SC (recovery time being directly dependent on the duration of the procedure). Pharmacological manipulation of in vivo spontaneous activity was achieved by re-anesthetizing animals using 3% isoflurane for 5–10 minutes, exposing the eye, and injecting ~500 nL per eye with a pulled glass micropipette. After injection the eyelid was closed and covered in ophthalmic ointment and a brief local anesthetic, and the animal returned to 0% isoflurane before being allowed to recover at least 20–30 minutes before resumption of imaging. Movies were analyzed using overlaid 10 μm square regions-of-interest (ROIs) as has been described previously (Ackman, 2012, also see Supplemental Experimental Procedures).

Wave Property Analyses

Inter-retina wave interval was measured by determining the difference between the time of the wave peak in opposing hemispheres in the same animal. Wave direction bias index was computed by grouping individual wave directions into 30-degree bins and calculating a vector sum of the angle and length of directions within each hemisphere. Recordings from GCaMP-loaded cells were preferred wherever possible for wave property analyses as cell loading was found to be much more uniform and consistent between hemispheres and animals.

Supplementary Material

1. Movie S1: Five minute SC retinal wave recordings from WT, β2^−/−, and β2^+/− retinal ganglion cell afferents. Related to Figure 1.

First panel is WT recording (ΔF/F in all panels) from both SC hemispheres showing clear, frequent wave-fronts. Second panel is from a young (P4) β2^−/− mouse and shows a lower frequency of waves (two in each hemisphere), along with bilateral wave correlation and decreased amplitude. Third panel is from a β2^+/− mouse, showing similar wave activity to WT, with frequent, high-amplitude waves with thin wave-fronts. Scale bars are 200 μm, 20x speed in all movies, which were acquired at 5 Hz.

2. Movie S2: SC wave recordings from WT and β2^−/− mice. Related to Figure 3.

First panel is P3 WT recording (ΔF/F) from SC right and left hemisphere. WT SC waves are similar to waves recorded from retinal ganglion cell axons in the SC in frequency and spatiotemporal properties. Second panel is P3 β2^−/− SC recording, showing frequent, high-speed, bursts of “small” wave activity near the midline, along with 1–2 waves that seem similar to waves recorded from β2^−/− RGC axons. Scale bars are 200 μm, 20x speed in all movies, which were acquired at 5 Hz.

3. Movie S3: SC wave recording from Pax6a-Cre⁺;β2^fl/+ (Retβ2-cHet) and Pax6a-Cre⁺;β2^fl/− (Retβ2-cKO) mice. Related to Figure 5.

First panel shows Retβ2-cHet animal activity (ΔF/F), with regions of Pax6a expression roughly represented by dotted lines. Regions within the medio-lateral dotted line area largely lack Pax6a-Cre, leading to the incomplete elimination of the floxed β2 subunit of the nAChR. This causes the medial region in the Retβ2-cKO (right panel) to retain one copy of the β2 subunit, while the rest of the retina acts as a β2-cKO. Similarly, in the Retβ2-cHet (left panel), the middle of the retina (and SC) retains both copies of β2, while the outer regions act as a β2-cHet. Wave activity in the Retβ2-cHet ignores the region boundaries and appears similar to WT waves (quantified in Figure 5), while Retβ2-cKO activity in the right panel is constrained within the middle region, while showing similar properties to WT waves, albeit truncated to the middle region of the SC (Figs 5, S5). Both movies are 20x speed, with 200 μm scale bars.

4. Movie S4: Whole-animal β2^−/− SC recording from RGC afferents before and after single-eye CPT-cAMP injection. Related to Figure 7.

Left panel is P4 β2^−/− RGC axon recording (same as in Movie S1) showing activity before CPT-cAMP injection. Right panel shows activity from the same animal roughly two hours after a ~2 mM CPT-cAMP injection (~500 nL) in the right eye. Retinal afferents in the contralateral left hemisphere of the SC show increased frequency of waves after CPT-cAMP injection, but the spatiotemporal properties of the induced waves appear identical to the altered β2^−/− retinal waves (quantified in Figs 2, 7).

5

Figure S1: Example β2^+/− Retinal Waves and Monte Carlo Analysis of Inter-retina Wave Correlation

A. Time-series montage of a single retinal wave in β2^+/− animal in one hemisphere. Scale bar = 200 μm, each frame is 1s apart from prior, P7. β2^+/− retinal waves closely resemble WT waves. Also see Movie S1. B. Raster plot of activity in regions of interest in β2^+/− retinal recordings, with points representing one 10×10μm region of interest. Waves in two hemispheres are frequent, similar to WT raster plots. C. Results of Monte Carlo analysis on WT, β2^−/−, and β2^+/− retinal wave recordings. Wave recordings were shifted randomly in one hemisphere and compared to the opposite, and the number of shifted 10-minute recordings with increased temporal correlation relative to the original was calculated (out of 1000 random shifts). Recordings in which less than 5% of shifts resulted in more correlation between hemispheres than the original were considered to demonstrate significant inter-retinal correlation (red). 28/38 (~74%) of β2^−/− movies were significantly correlated, while only 1/22 (~5%) of WT or β2^+/− recordings were correlated between retinas.

Figure S2: Effects of Meclofenamic Acid (MFA) and Enucleation on Retinal Waves in β2^+/− Mice.

A. Method for pharmacological manipulation of retinal waves in β2^−/− retinas. Undisturbed waves were recorded from both hemispheres before single eye injection of Meclofenamic acid (MFA, 4–20mM, Sigma). B. Raster of β2^−/− retinal wave activity before (top) and after (middle) single (left) eye injection of MFA. Waves are eliminated in the SC hemisphere contralateral to the eye that received the MFA injection in a P7 mouse. Bottom panel shows β2^−/− retinal wave activity after monocular enucleation. Waves are eliminated in the hemisphere contralateral to the enucleation in a P6 mouse. C. Quantification of MFA and enucleation effects on wave frequency, by hemisphere. Waves are decreased in frequency in the hemisphere contralateral to the MFA injected (top) and enucleated (bottom) eyes (Ages = P5–7). D. Wave size in WT, β2^−/− and β2^+/−. β2^−/− waves are larger than WT waves. E. Region of interest (ROI) activation frequency is higher in WT than it is in β2^−/−. F. Region of interest activation (inter-wave) interval. Activation interval is higher in β2^−/− hemispheres than in WT. G. Region of interest signal amplitude during waves. WT ROI wave amplitude is higher than in β2^−/− mice. H. ROI activation duration is indistinguishable between WT and β2^−/− mice.

Figure S3: Further Properties of ‘Small’ SC Waves in β2^−/− Mice

A. Single wave montage of β2^−/− SC “small” wave with acquired with a high (40 Hz) frame rate (100 ms/panel shown). Montage demonstrates that small waves do propagate, and that they can move across the anterior-posterior length of the SC within 200–300ms, at much higher speeds than either WT or β2^−/− retinal waves. B. Wave duration by hemisphere. Waves are shorter in duration in β2^−/− SC recordings than in β2^−/− Retinal recordings. WT SC waves are longer in duration that both WT retinal waves and β2^−/− SC waves. C. Wave size by hemisphere. Wave size is larger in WT SC recordings than retinal recordings, but β2^−/− SC waves are smaller than β2^−/− retinal waves. D. Wave speed by hemisphere. Wave speed is higher in β2^−/− SC recordings than in β2^−/− retinal and WT recordings. E. Wave direction plots. WT SC wave direction is similar to WT retinal wave direction bias, but β2^−/− SC bias is retained, but bipolar in contrast to the unipolar β2^−/− retinal wave direction bias.

Figure S4: β2 Knockout Creation, Anatomy.

A. Method for creation of floxed chrnb2 gene (also see Experimental Procedures and Supplemental Experimental Procedures). B. Retinotopic map refinement is disrupted at P8 in germline deletion knockout, similarly to another germline β2^−/− null mouse (Picciotto) in both dorsal (C.) and ventral (D.) RGCs. E. Eye-specific segregation is disrupted in the germline β2^−/− mouse in both SC and dLGN (H) in both fraction of misprojected ipsilateral axons in the contralateral region (F, I) and overlap of ipsilateral and contralateral axons (G, J).

Figure S5: SC β2^−/− Conditional Knockout Wave Properties, Anatomy.

A. Wave frequency plot by hemisphere in WT, β2^−/−, and SCβ2-cKO (En1-Cre⁺;β2^fl/fl) mice. Wave frequency in SCβ2-cKO is comparable to WT frequency, lower than whole-animal β2^−/−. B. Wave duration by hemisphere. Wave duration in SCβ2-cKO is not different from WT and is larger than β2^−/−. C. Wave speed in SCβ2-cKO is not different from WT wave speed, and is less than whole-animal β2^−/−. D. Eye-specific segregation in En1-Cre⁻;β2^fl/− (another littermate control) mouse SC is normal at P8. E. Retinotopic mapping in the En1-Cre^−;β^2fl/− (another littermate control) mouse SC is also normal at P8.

Figure S6: Retβ2-cKO Wave Duration and In Vitro Wave Activity.

A. Wave duration plotted by hemisphere in WT, β2^−/−, Retβ2-cKO and Pax6α-Cre⁺;β2^fl/+ (Retβ2-cHet) mice. Wave duration in Retβ2-cKO is smaller than WT and Retβ2-cHet mice. B. In vitro wave frequency and firing rate in medial region of Retβ2-cKO and Retβ2-cHet. Wave frequency and firing rate of RGCs examined in vitro with a multielectrode array are unchanged within the medial, Cre-spared region of the retina. Approximate multielectrode array recording site noted in diagram. C. Retinotopic mapping in dLGN in WT, Retβ2-cKO and Retβ2-cHet mice at P8 assayed using focal DiI injections into ventro-temporal (VT) or dorsal retina. Retinotopy of ventro-temporal projections are disrupted in Retβ2-cKO mice (in Cre+ region), but are not disrupted for dorsal projections (in largely Cre- region).

Figure S7: CPT-cAMP Effects on β2^−/− Retinotopy in SC, Wave Direction Bias, and Inter-retinal Wave Interval Correlation

A. Experimental procedure for imaging retinal wave activity before and after single-eye injection of CPT-cAMP. B. Raster plot of five-minute recordings of calcium dye signal from retinal ganglion cell axons before (top) and after (bottom) single RT-eye injection of CPT-cAMP. Wave frequency is significantly increased only contralaterally to the injected eye (also see Fig. 7C) C. Wave direction bias in β2^−/− mice after CPT-cAMP in both retinal wave and SC wave recordings. Waves are similarly biased as in β2^−/− recordings before cAMP (right histogram), and retinal and SC cAMP wave directionality is comparable. C. Inter-retina wave interval correlation in β2^−/− after binocular CPT-cAMP. Correlation between waves in the two hemispheres remains high after binocular application of CPT-cAMP in β2^−/− mice. Monte Carlo analysis results of inter-retina wave correlation after binocular CPT-cAMP. 7/8 (~88%) of 10-minute recordings are significantly correlated after binocular CPT-cAMP injections. E. Threshold-dependent analysis of eye-specific segregation effects in β2^−/− mice after five days of binocular cAMP injections (P2–6). In both the SC (top) and dLGN (bottom) there is a partial rescue of eye-specific segregation across multiple thresholds.