Abstract
Article Figures and data Abstract Introduction Results and discussion Materials and methods References Decision letter Author response Article and author information Metrics Abstract Neurons receive synaptic inputs on extensive neurite arbors. How information is organized across arbors and how local processing in neurites contributes to circuit function is mostly unknown. Here, we used two-photon Ca2+ imaging to study visual processing in VGluT3-expressing amacrine cells (VG3-ACs) in the mouse retina. Contrast preferences (ON vs. OFF) varied across VG3-AC arbors depending on the laminar position of neurites, with ON responses preferring larger stimuli than OFF responses. Although arbors of neighboring cells overlap extensively, imaging population activity revealed continuous topographic maps of visual space in the VG3-AC plexus. All VG3-AC neurites responded strongly to object motion, but remained silent during global image motion. Thus, VG3-AC arbors limit vertical and lateral integration of contrast and location information, respectively. We propose that this local processing enables the dense VG3-AC plexus to contribute precise object motion signals to diverse targets without distorting target-specific contrast preferences and spatial receptive fields. https://doi.org/10.7554/eLife.31307.001 Introduction Neurons receive most of their synaptic input on large intricately branched dendritic arborizations. Traditionally, distributed inputs were thought to be summed linearly at the cell body (Yuste, 2011). However, recent studies uncovered extensive local processing and clustered plasticity of synaptic inputs, which enhance the computational power of dendrites (Grienberger et al., 2015; Harvey and Svoboda, 2007; Kleindienst et al., 2011; London and Häusser, 2005; Losonczy et al., 2008). Although less studied, similar local processing occurs in terminal axon arbors, in which presynaptic inhibition and inhomogeneous distributions of voltage-gated ion channels can diversify the output of a single neuron (Debanne, 2004; Asari and Meister, 2012). Amacrine cells (ACs) are a diverse class of interneurons in the retina (Helmstaedter et al., 2013; MacNeil and Masland, 1998). Most of the approximately 50 AC types lack separate dendrites and axons and receive input and provide output through the same neurites (Diamond, 2017). Among the few AC types that have been studied in detail, starburst and A17 ACs are critical for direction selectivity and dim light signaling, respectively (Grimes et al., 2015; Amthor et al., 2002; Vlasits et al., 2014; Yonehara et al., 2016; Yoshida et al., 2001). The radially symmetric arbors of starburst ACs receive synaptic input and release neurotransmitters near and far from the soma, respectively (Ding et al., 2016; Vlasits et al., 2016). In a seminal study, Euler et al. (2002) discovered by two-photon Ca2+ imaging that the four to six primary neurites of starburst ACs with their daughter branches function as independent centrifugal motion sensors. Similarly, A17 ACs were shown to process converging inputs from rod bipolar cells separately (Grimes et al., 2010). For most AC types, however, whether arbors process inputs locally or integrate them globally and what specific stimulus features neurites encode remains unknown. As in most parts of the nervous system, synaptic communication in the retina occurs in dense neuropils in which arbors of neighboring cells overlap extensively (Helmstaedter et al., 2013). Population coding in sensory and motor systems has been studied at the level of cell bodies (Arnson and Holy, 2013; Churchland et al., 2012; Leonardo and Meister, 2013), but how cell-type-specific information is organized in population activity in neuropils has not been explored. VG3-AC neurites stratify broadly in the center of the inner plexiform layer (IPL) forming a dense plexus in which processes of approximately seven cells overlap at any point (Haverkamp and Wässle, 2004; Johnson et al., 2004; Kim et al., 2015). In somatic patch clamp recordings, VG3-ACs depolarize to light increments (ON) and decrements (OFF) restricted to their receptive field center, but hyperpolarize to large ON and OFF stimuli that include their receptive field surround (Kim et al., 2015; Lee et al., 2014; Grimes et al., 2011). In addition, VG3-ACs depolarize strongly to local object motion but hyperpolarize during global image motion as occurs during eye movements (Kim et al., 2015). VG3-ACs are dual transmitter neurons. They provide glutamatergic input to a group of motion-sensitive retinal ganglion cell (RGC) types with diverse contrast and stimulus-size preferences (Krishnaswamy et al., 2015; Kim et al., 2015; Lee et al., 2014), and provide glycinergic input to Suppressed-by-Contrast RGCs (SbC-RGCs), inhibiting selectively responses to small OFF stimuli (Lee et al., 2016; Tien et al., 2016; Tien et al., 2015). Whether VG3-AC neurite arbors process inputs locally or integrate them globally, what stimulus features they encode, and how visual information is organized in the population activity of the VG3-AC plexus to support its varied circuit functions is unknown. Here, we used two-photon Ca2+ imaging in a novel transgenic mouse line to address these questions. Results and discussion We crossed VG3-Cre mice to a novel transgenic strain (Ai148) expressing the genetically encoded Ca2+ indicator GCaMP6f in a Cre-dependent manner enhanced by tTA-based transcriptional amplification. Staining for VGluT3 confirmed that GCaMP6f labeling in VG3-Cre:Ai148 retinas was mostly restricted to VG3-ACs (Figure 1—figure supplement 2) with sparse off-target expression in RGCs (Grimes et al., 2011; Kim et al., 2015). We imaged GCaMP6f signals in scan fields (33 × 33 μm for Figures 1, 2 and 4; 13 × 100 μm for Figure 3) in the IPL of flat-mounted retinas at 9.5 Hz with a pixel density of 4.7 pixels / μm2. Recording depths of scan fields were registered by their relative distance to the outer and inner boundaries of the IPL (0–100%) detected by imaging transmitted laser light (Figure 1—figure supplement 2). Visual stimulation (385 nm) was spectrally separated from GCaMP6f imaging (excitation: 940 nm, peak emission: 515 nm); and recordings were obtained from the ventral retina, where S-opsin dominates (Haverkamp et al., 2005; Wang et al., 2011). To objectively identify processing domains of VG3-ACs neurites, we segmented images into functionally distinct regions of interest (ROIs) using a serial clustering procedure (Figure 1—figure supplement 1; s. Materials and methods). Figure 1 with 3 supplements see all Download asset Open asset Contrast preferences of VG3-AC neurites shift across IPL depths. (A) Schematic of the VG3-AC circuit. VG3-AC neurites receive input from ON and OFF bipolar cells (ON and OFF BC) and synapse onto RGCs. (B) Ca2+ transients of ROIs at different imaging depth elicited by contrast steps in a small spot (diameter: 100 μm). A bar at the top indicates the stimulus timing. The black trace (shaded area) shows the mean (±SEM) responses of VG3-AC somata (n = 15). The six color-coded traces (shaded areas) indicate the mean (±SEM) responses of neurite ROIs at different IPL depths (21%: n = 673, purple; 29%: n = 972, blue; 37%: n = 817, sky; 44%: n = 1029, green; 51%: n = 1380, lime; 60%: n = 928, olive). (C, D) Distributions (C) and mean ± SEM (D) of polarity indices of VG3-AC neurite ROIs at different IPL depths color-coded as in (B). Polarity indices differed between IPL depths (p<10−16, Kruskal-Wallis one-way ANOVA). ROIs at 21% and 29% IPL depth were more biased to OFF responses than at other depths (p<10−4 compared to 37%; p<10−7 for 44–60%). ROIs from 51–60% IPL depth were more biased to ON than ROIs from 21–44% (p<10−7). No significant differences were observed between 21% and 29% (p=0.99) and between 51% and 60% (p=0.98). Even without image segmentation, using the average activity of each image plane a single data point, polarity indices differed across IPL depths (p<10−12, Kruskal-Wallis one-way ANOVA; 21%: n = 15; 29%: n = 18; 37%: n = 14; 44%: n = 16; 51%: n = 23; 60%: n = 20). (E) Lines show the distributions (i.e. skeleton densities) of axons of different OFF (BC1 – BC4) and ON (BC5t – BC7) bipolar cells types from 15–65% IPL depth, according to (Greene et al., 2016; Helmstaedter et al., 2013) (F) Representative image of a VG3-AC filled with Oregon Green BAPTA-1 via a patch-clamp pipette. (G, H) The average responses (±SEM, G) and polarity index distributions (H) of ROIs of a single VG3-AC at two IPL depths (34%: n = 50, blue; 47%: n = 59, green). (I) Depth-dependent shift in polarity indices (mean ±SEM) of neurite ROIs of two VG3-ACs filled with Oregon Green BAPTA-1 (depth-dependent differences within cells p<10−8 and p<0.05). https://doi.org/10.7554/eLife.31307.002 Figure 2 with 3 supplements see all Download asset Open asset ON and OFF responses in VG3-AC neurites differ in preferred stimulus size, but are equally transient. (A) Ca2+ responses of ROIs at different imaging depths to contrast steps in spots of different size. Spot diameters are noted above the bars indicating stimulus timing. The black traces (shaded areas) show the mean (±SEM) responses of VG3-AC somata (n = 8). The color-coded traces (shaded areas) indicate the mean (±SEM) responses of ROIs at different IPL depths (21%: n = 306, purple; 29%: n = 456, blue; 37%: n = 336, sky; 43%: n = 367, green; 50%: n = 700, lime; 60%: n = 588, olive). (B) The distributions of ON (white) and OFF (dark gray) receptive field center sizes of VG3-AC neurite ROIs. ON receptive field centers were larger than OFF receptive field centers (p<10−99, Wilcoxon rank sum test) (C) Receptive field center sizes (mean ±SEM) of ON (open circles) and OFF (filled circles) responses as a function of IPL depth. Because small response amplitudes led to rejection of >50% of ON responses of ROIs from 21–37% IPL depth (s. Material and methods), we restrict comparisons to 43–60% IPL depth. At all these depths, ON receptive field center sizes were larger than OFF receptive field center sizes (43%: p<10−8, 50%: p<10−9, 60%: p<10−9, Wilcoxon rank sum test with multiple comparison correction using the Benjamini–Hochberg procedure). Even without image segmentation, using the average activity of each image plane a single data point, ON receptive field centers were larger than OFF receptive field centers (p<10−3, Wilcoxon rank sum test, total: n = 61, 21%: n = 6; 29%: n = 13; 37%: n = 7; 43%: n = 6; 50%: n = 12; 60%: n = 17). (D) The distributions of transience indices of ON (white) and OFF (dark gray) responses of VG3-AC neurite ROIs did not differ significantly (p=0.925, Wilcoxon rank sum test). (E) Summary data (mean ± SEM) of transience indices of ON (open circle) and OFF (filled circle) responses as a function of IPL depth. Due to the high rejection rate of ON responses from 21 to 37% IPL depth, comparisons were restricted to 43–60% IPL depth. Transience indices of ON response is marginally lower than those of OFF responses at 43% (p<0.05), but were not significantly different at 50% (p=0.82) and 60% (p=0.05) IPL depth (Wilcoxon rank sum test with multiple comparison correction using the Benjamini–Hochberg procedure). We are not sure what accounts for the greater response transience observed in VG3-AC neurites vs. somata (p<10−4, Wilcoxon rank sum test). One possibility is that inhibitory synaptic inputs favor neurites and abbreviate responses. https://doi.org/10.7554/eLife.31307.006 Figure 3 Download asset Open asset Population activity of the VG3-AC plexus encodes spatial information with high precision. (A) Schematic of visual stimulus. Vertical bars (height: 60–80 μm, height: 50 μm width) were presented at 17 different positions along the horizontal axis of a rectangular imaging region (height: 13 μm, width: 100 μm). Stimulus positions were symmetric around the center of the imaging region and spaced by 25 μm center-center distances from −150 μm to 150 μm. In addition, bars were presented −400 μm, −200 μm, 200 μm, and 400 μm from the center of the imaging region. Each bar was presented three for 1.5 s with an interval of 1.5 s between stimuli. The order of stimulus positions was randomized and each stimulus repeated three times. (B) Heatmap of normalized responses in VG3-AC neurites to bars stimuli from −150 μm to 100 μm from the center of the imaging region at an IPL depth of 53%. Responses have been reordered by stimulus positions. Each row of the heatmap represents the activity a single pixel. Pixels were sorted by their distance from the center of the imaging region (−50 μm to 50 μm). (C) Representative images of the VG3-AC plexus in the scan region obtained by averaging the GCaMP6f signal over (left: IPL depth 36%, right: IPL depth 53%). (D) Maps of receptive field positions in the same regions of the VG3-AC plexus shown in (C) (left: OFF responses, right: ON responses). (E) Distributions of receptive field positions of pixels in four adjacent subsections (color-coded from left to right in: purple, sky, lime, and olive) of the scan regions shown in (C) and (D). Receptive field positions of the pixels of each image were aligned to zero their average. (F, H) The accuracy with which a naive Bayes classifier can assign the location of a VG3-AC neurite pixel based on its receptive field position to one of two image subsections increases as a function of the distance between these subsections (solid lines). Dashed lines shows the accuracy when classifiers were trained on shuffled receptive field positions. (G, I) The minimum separable distance (i.e. the point at which prediction accuracy reaches 75%) decreased when predictions were based on multiple pixels (e.g. median ROI size in VG3-AC neurites: 10 pixels). https://doi.org/10.7554/eLife.31307.010 Figure 4 with 1 supplement see all Download asset Open asset Uniform local motion preference of VG3-AC neurites. (A) Schematic at the top shows the time course of the grating motion in the receptive field center and surround (s. Video 2, and Materials and methods). The black trace (shaded area) shows the mean (±SEM) responses of VG3-AC somata (n = 11). The color-coded traces (shaded areas) indicate the mean (±SEM) responses of ROIs at different IPL depths (24%: n = 388, purple; 29%: n = 202, blue; 36%: n = 500, sky; 43%: n = 322, green; 49%: n = 308, lime; 58%: n = 298, olive). (B) The distribution of local motion preference indices of all ROIs. (C) Summary data (mean ± SEM) of local motion preference indices as a function of IPL depth. Local motion preference indices did not differ across IPL depths (p=0.09, Kruskal-Wallis one-way ANOVA). No ROI group at any depth was significantly different from any ROI group at another depth. https://doi.org/10.7554/eLife.31307.011 In somatic patch clamp recordings, VG3-ACs depolarize to small ON and OFF stimuli (Lee et al., 2014; Kim et al., 2015; Grimes et al., 2011). Somatic Ca2+ transients exhibited similar ON-OFF profiles (Figure 1A and B). To test how ON and OFF responses are distributed across VG3-AC arbors, we recorded Ca2+ transients elicited by contrast steps in a small spot (diameter: 100 μm) at different depths of the IPL (Figure 1B and Video 1). We quantified contrast preferences by a polarity index, ranging from −1 for pure OFF responses to 1 for pure ON responses (see Materials and methods). Polarity indices varied widely between ROIs (n = 5814, n = 11 mice). The distribution of polarity indices shifted with IPL depth, as neurites in the outer IPL (depths < 40%) responded more strongly to OFF stimuli, and neurites in the inner IPL (depths > 40%) responded more strongly to ON stimuli (Figure 1C,D). To make sure that the sparse off-target expression of GCaMP6f in RGCs did not contribute significantly to these results, we imaged signals in the IPL of VG3-Cre:Ai148 mice 3 weeks after optic nerve crush, which removes most RGCs but not ACs (Park et al., 2008). Distributions of polarity indices measured in these experiments recapitulated the depth-dependent shift in contrast preferences observed in control retinas (Figure 1—figure supplement 3). Because the arbors of each VG3-AC span the depth of the VG3-AC plexus (Grimes et al., 2011; Kim et al., 2015; Lee et al., 2014), it seemed unlikely that the shift in contrast preferences reflected differences between cells. Nonetheless, we imaged Ca2+ transients in two VG3-ACs filled with Oregon Green BAPTA-1, confirming that polarity indices shift within arbors of single cells (Figure 1F–I). The ratio of ON and OFF signals across VG3-AC arbors closely followed stratification patterns of ON and OFF bipolar cell axons in the IPL (Figure 1E) (Helmstaedter et al., 2013; Franke et al., 2017; Greene et al., 2016). However, response polarities of VG3-AC neurites were less extreme than those reported for bipolar cell axons (Borghuis et al., 2013; Franke et al., 2017). This suggests that local bipolar cell innervation patterns and restricted postsynaptic signal (voltage and/or Ca2+) spread determine contrast preferences of VG3-AC neurites and limit vertical integration of visual information in their arbors. Video 1 Download asset This video cannot be played in place because your browser does support HTML5 video. You may still download the video for offline viewing. Download as MPEG-4 Download as WebM Download as Ogg Ca2+imaging of VG3-AC neurite responses to contrast steps in spots of varying size recorded at different IPL depths. Image series of GCaMP6f responses at 24% (middle) and 53% (right) IPL depth to contrast steps in spots of different size (left). The video is sped up 2.5-fold relative to the image acquisition. In the left panel, the area of the scan fields is indicated by a red box. Two average normalized ROI traces are shown at the bottom of the middle and the right panel. https://doi.org/10.7554/eLife.31307.013 A hallmark of VG3-ACs’ somatic voltage responses is strong size selectivity (Lee et al., 2014; Kim and Kerschensteiner, 2017; Kim et al., 2015). We therefore explored how VG3-AC neurites respond to contrast steps in spots of different sizes (Figure 2A and Video 1). The depth-dependent shift in contrast preferences of VG3-AC neurites observed for 100 μm spots persisted when we calculated polarity indices based on responses to all stimulus sizes (Figure 2—figure supplement 1). At all depths, only small stimuli (diameter <400 μm) elicited Ca2+ transients in VG3-AC neurites (Figure 2A and Video 1) and size-selectivity indices of ROIs were uniformly high (Figure 2—figure supplement 2), indicating that receptive field surrounds are strong across VG3-AC arbors. To measure ON and OFF receptive field centers, we estimated optimal stimulus sizes for each ROI using a template-matching algorithm (see Materials and methods). ON receptive field centers of VG3-AC neurites were consistently larger than OFF receptive field centers, independent of IPL depth (Figure 2A–C). This could be due to larger dendritic territories of the ON compared to the OFF bipolar cells that provide input to VG3-ACs (Behrens et al., 2016), and/or the fact that ON but not OFF bipolar cell axons are gap junctionally coupled to AII ACs (Marc et al., 2014; Demb and Singer, 2015; Bloomfield and Völgyi, 2009). Both ON and OFF receptive field centers were smaller than VG3-AC arbors and only slightly larger than bipolar cell receptive field centers (Franke et al., 2017; Schwartz et al., 2012; Purgert and Lukasiewicz, 2015), supporting the notion that local input from a small number of bipolar cells shapes spatial receptive fields of VG3-AC neurites with limited lateral integration of visual information in their arbors. In contrast to differences in their spatial tuning, ON and OFF responses were equally transient across VG3-AC arbors (Figure 2A,D,E). By increasing our scan rate from 9.5 Hz to 37.9 Hz, we confirmed that our measurements of response transience were not limited by the image acquisition rate (Figure 2—figure supplement 3). At any point of the VG3-AC plexus, arbors from approximately seven cells overlap (Kim et al., 2015). To explore how spatial information is encoded by population activity in this plexus, we imaged rectangular regions (height: 13 μm, width: 100 μm) in the IPL of VG3-Cre:Ai148 mice bars (height: 60–80 μm, width: 50 μm) at different positions 25 μm, μm) along the horizontal axis of the imaging region. We presented each bar for 1.5 s with an interval of 1.5 s between stimuli. were shown in and responses reordered by stimulus positions in Figure and Video We ON and OFF responses but data from different IPL depths, which did not differ in their spatial coding (Figure 2). For each we receptive field positions along the horizontal stimulus axis (Figure see Materials and methods). This revealed continuous topographic maps of visual space in the VG3-AC plexus (Figure To the of these we calculated the accuracy with which naive Bayes classifiers could assign neurite activity to specific parts of the based on receptive field positions (see Materials and methods). Even for single this accuracy was high (Figure and the distance at which different regions of the could be with accuracy (i.e. separable decreased when that multiple pixels the activity of VG3-AC neurite processing domains number of pixels Figure Thus, local processing precise topographic maps of visual space in the population activity of the dense VG3-AC plexus. Video 2 Download asset This video cannot be played in place because your browser does support HTML5 video. You may still download the video for offline viewing. Download as MPEG-4 Download as WebM Download as Ogg Ca2+imaging of VG3-AC neurite responses to bar at different distance from the Image series of GCaMP6f responses at 53% IPL depth to a vertical stimulus bar s 1.5 s OFF) presented at different positions along the horizontal axis of the imaging region. The video is sped up 2.5-fold relative to the image acquisition. The scan field is indicated by a red box. response traces of pixels from four different parts of scan field are shown at the color-coded according to the bar the scan VG3-ACs in in the retina (Krishnaswamy et al., 2015; Kim et al., 2015; Kim and Kerschensteiner, 2017). We the of VG3-AC neurites to local and global image motion, using a stimulus in which center and surround regions of receptive fields separately or (Kim et al., 2015; et al., et al., 2012). At all IPL depths, motion of the center grating elicited Ca2+ transients in VG3-AC neurites, which remained silent during motion of in center and surround (i.e. global (Figure Figure supplement 1, and Video 3). As a local motion preference indices (see Materials and of of ROIs were (Figure Thus, in of the of responses to contrast VG3-AC neurites object motion Video 3 Download asset This video cannot be played in place because your browser does support HTML5 video. You may still download the video for offline viewing. Download as MPEG-4 Download as WebM Download as Ogg Ca2+imaging of VG3-AC neurite responses to motion. Image series of GCaMP6f responses at IPL depth (right) to or motion of in the center and surround separated by a (left). The video is sped up relative to the image acquisition. In the left panel, the area of the scan fields is indicated by a red box. The average normalized ROI trace is shown at the bottom of the right panel. How does local processing in neurites of VG3-ACs contribute to their circuit VG3-ACs provide glutamatergic input to which movements in a small area of visual space closely aligned with their dendrites (Kim et al., 2015; et al., 2012). from VG3-ACs is for responses of (Kim et al., 2015). VG3-ACs visual information globally, receptive fields of the with which the position of could be from their activity and 2017). In to VG3-ACs provide input to ON ganglion cells (ON ON-OFF and (Lee et al., 2014; et al., 2015). motion-sensitive types differ in their preferred stimulus contrast and stratify dendrites at different depths of the The depth-dependent shift in contrast preferences across neurite arbors enables VG3-ACs to contribute motion-sensitive input to ON ON-OFF and without the diverse contrast preference of these VG3-ACs provide glycinergic input to et al., 2016; Tien et al., 2015; Lee et al., 2016). Whether VG3-ACs release and from different in their and how these differ in their visual information remains to be Nonetheless, when VG3-ACs were from the retina, inhibitory input to was for OFF but not ON stimuli et al., 2016). The local processing of ON and OFF signals we in VG3-AC arbors could this we because of local population activity in the VG3-AC plexus local presynaptic input patterns than postsynaptic and represents visual space in precise continuous topographic Thus, local processing enables the dense VG3-AC plexus to contribute precise and uniformly object motion signals to diverse targets without distorting target-specific contrast preferences and spatial receptive fields. Materials and methods We crossed VG3-Cre in which is from a of the by (Grimes et al., to the a novel transgenic line by a into the on followed by of that by mice within the et al., for GCaMP6f and for tTA-based transcriptional of GCaMP6f in a two mouse To of VG3-ACs two-photon for with Oregon Green BAPTA-1, we crossed VG3-Cre mice to the strain et al., 2010). were in a and We retinas from mice of between and All in this study were by the and of of and were in with the of for the and of nerve a were with a of and The optic nerve was and with for the of the At the of a of was for and was and after was to the for a were for more than 1 with by and were in mouse with for or for two-photon 1 11 and to with 1 2 and with retinas were on black for or for two-photon a retinas were for in in at and three for