All procedures were performed at Northwestern University and approved by the Institutional Animal Care and Use Committee. Retinas were obtained from both male and female thirteen-lined ground squirrels (Ictidomys tridecemlineatus) in approximately equal numbers. Retinal slices, 100 µm thick, were obtained using a razor mounted on vertical slide from ~2 × 2 mm squares mounted vitreal side down on Millipore filter paper35,78. For experiments involving light responses, retinal slices were obtained under dim red illumination79. During recordings, slices were visualized with a Zeiss Axioskop-2FS microscope using a 63x water immersion objective under infra-red illumination. Recordings were made with Axopatch 200B amplifiers (Molecular Devices) and signals were filtered at 5 kHz and digitized at a rate of 10 or 16.6 kHz with a HEKA ITC-18 A/D board (HEKA Elektronik) operated with custom software (Igor Pro 6.21; WaveMetrics). Patch pipettes were pulled from borosilicate glass capillary tubes to tip resistances of 8-12 MΩ.
The external solution consisted of (in mM): NaCl 115, KCl 3.1, MgSO4 2.48, glucose 6, Na-succinate 1, Na-malate 1, Na-lactate 1, Na-pyruvate 1, CaCl2 2, and NaHCO3 25, 0.05% phenol red, and was equilibrated with 5% CO2/95% O2 to a pH of 7.4. Picrotoxin (50 µM; Sigma, P1675) and strychnine (10 µM; Sigma, S0532) were included in the bath. The patch pipette solution for the BC contained (in mM): KCl 120, K3-EGTA 10, MgSO4 2, HEPES 10, ATP 5 and GTP 0.5; pH 7.35 with KOH. The cone pipette solution contained (in mM): KSCN 115, K3-EGTA 10, MgSO4 2, HEPES 20, ATP 5 and GTP 0.5; pH 7.35 with KOH. Intracellular solutions contained combinations of a non-fixable tracer, either Sulforhodamine 101 (Molecular Probes, S359) or BODIPY 492/515 (Molecular Probes, D3238) and a fixable tracer, either Cascade blue hydrazide, trilithium salt (0.1 mM; Invitrogen, C3239) or 0.5% Neurobiotin Tracer (Vector Laboratories, SP-1120). Labeled cells could be visualized using immunohistochemistry followed by confocal imaging6 or immediately following recording with a Prime95B (Photometrics) camera controlled by µManager (micro-manager.org). All solutions were corrected to an milliosmolarity of 285 ± 5. Experiments were performed at 32-33 oC. Pharmacological agents include: UBP310 (Tocris, #3621), GYKI53655 HCl (Tocris, #2555), DL-TBOA (Tocris, #1223), dihydrokainic acid (Tocris, #0111), and glutamic acid (Sigma, G1251). Membrane capacitance was measured with a HEKA EPC-10 using the ‘sine+dc’ Lockin routine in the Patchmaster software (2×73.3). For capacitance measurements, CsCl 112 and KCl 8 mM were substituted for KCl in the intracellular solution and solution pH was adjused with CsOH18. Retinas were stimulated by a light-emitting diode (574 nm) attached to a microscope video port. LED intensity was controlled by pulse-width modulation. Light sources were calibrated with a photodiode detector (International Light) that was positioned beneath the microscope objective. Light intensity was converted to photons at 520 nm, the λmax for the ground squirrel green cone pigment80. Figure legends show light intensities. During light stimuli, unitary response amplitude and number were analyzed using Campbell’s theorem60,
where and are the signal mean and variance and the integrals are equal to event area and event area squared. Canonical events were obtained from cb1 and cb2 cells during 1 ms cone depolarizations. The event amplitude was scaled to find the unique ratio that matched the variance to mean ratio. Application of Campbell’s theorem assumes both linear event summation and that the predominant source of shot noise doesn’t change with light step intensity. Both conditions may be approximate during BC light responses over a range of intensities. Variance and mean measurements after a strong flash (Fig. 9a) were obtained after subtracting a sloped, linear baseline.
Epsc and tpsc traces were Gaussian filtered in software with cutoffs of 1000–1250 Hz and 500-750 Hz, respectively, before further processing. Cones were stepped in voltage clamp from a holding potential of –70 mV to a level between –50 and 0 mV for 1 ms. Slow tpsc responses during an epoch (a train of steps to the same voltage) were measured as the average current during a 5 ms window that encompassed the inward current maximum. Average baseline current in an 8 ms window prior to the step was subtracted from the maximum. Peak responses from one or more consecutive epochs were accumulated into amplitude histograms. Amplitude histograms were fitted by a sum of Gaussians with variables for individual peak amplitudes (including failures), mean quantal size (Δx), and failure and success peak Gaussian widths, σf and σs, respectively. The peak location of the failure distribution, x0, was unconstrained. The Gaussian widths for the failure and subsequent success peaks followed the formula (σf2 + n(σs2))1/2 where n is peak number. Each tpsc response from an entire run consisting of multiple epochs was assigned an effective quantal content, m, by binning according to x0 + Δx(m – 1/2), x0 + Δx(m + 1/2). Tpscs with m ≤ 12 were assumed to sum linearly and constituted the cone-linear range in scatter plots. Assignments were inspected for the presence of spontaneous cone transporter events either before or immediately after the 1 ms stimulus. In the case of overlap, amplitudes were either assigned manually or, in rare cases where assignment remained uncertain, the trial was discarded.
Epsc peak amplitude was determined for all Off BC types using automated routines35. In brief, copies of the “original” traces were filtered in software with a Gaussian filter (cutoff frequency 200 Hz). The baseline noise was calculated from time zero to the start of the 1 ms step for each filtered trace and the values for all individual traces were averaged to obtain σn. The response threshold was empirically set to 3.5σn, which was typically 0.8–1.5 pA. Each 200 Hz filtered trace was then sorted into one of two groups, failures or successes, based on whether the epsc exceeded threshold during the response interval which was determined by taking an average over all traces. Next, to determine the event shape, all original epsc responses in the series were averaged and the result fitted by least squares minimization with a function that could approximate both fast and large cb2 and slow and small cb1/3 events. The shape of the function, ConvExpDiffusion, was determined by convolving a fast exponential rise (τ = 0.1 ms) and decay (τ = 1.0 ms) with a profile obtained from an equation for diffusion from a point source in three dimensions. The values from the diffusion equation for peak amplitude, diffusion radius, and start time were saved. For responses in the ‘success’ group, peak amplitude, start time, and rise time were determined for each original trace by fitting with ConvExpDiffusion. For responses in the ‘failure’ group, original traces were fitted within the average response shape that was scaled by amplitude to produce a minimization of the sum of squares. Amplitudes could be positive or negative. The fitting routine issued error messages for instances of poor fit, which could then be inspected for interference by spontaneous events or noise. The fitting routine was robust to small cb1 and cb3 cell epscs whose rise and peak regions were often interrupted by apparent single channel transitions.
For plots of percent failure versus cone quantal content, the confidence interval for each point was based on a maximum likelihood estimation for coin tosses. For a biased coin with an intrinsic probability of heads, p, the binomial distribution gives the probability of observing a specific number of heads, h, in n trials. The cone to Off BC synapse presents the inverse problem where h (i.e., the number of failures) and n, the number events with a specific tpsc quantal content, are known, and the task is to estimate the likelihood that the intrinsic probability, p, is within a certain range of values. The relative likelihood of an intrinsic probability p given n and h is obtained from the beta probability distribution with parameters h + 1 and n – h + 1 such that the greatest likelihood occurs at p = h/n and the confidence interval is bounded by the probability values that exclude the lower and upper 25% of the area under the curve, respectively (i.e., the 50% confidence interval). Sample sizes of 2 or smaller were omitted from plots due to the large uncertainty associated with the measurement.
A rapid perfusion pipette was mounted on a piezoelectric actuator (Burleigh, PZS-200) and driven by an amplifier (Burleigh, PZ-150 M). The command voltage for the amplifier step, typically 18 or 60 ms in duration, was filtered in software by convolution with a Gaussian (width = 1.2 ms) to damp oscillations. Four-barreled glass (Vitrocom, Mountain Lakes, NJ) was mounted on a Kopf vertical puller and pulled so that each square barrel had a width of ~100 µm. The two side barrels were sealed with Sylgard. Solutions were fed under pressure to the two central barrels via small bore polyimide-coated quartz tubes (Polymicro Technologies, Phoenix, AZ). The flow into one barrel contained control solution and had a single input. The flow into the other barrel could be switched among five test solutions. Control solution contained (in mM): NaCl 125, KCl 3.1, MgSO4 1.24, CaCl2 2, and Na-HEPES 10. pH was adjusted to 7.4 with NaOH and osmolarity was adjusted to 285 ± 5 mosm with NaCl. Na-glutamate (18 mM) was substituted for NaCl to make the stock test solution. Serial glutamate dilutions were made by combining stock and control solutions. Junction potential differences between the control and 18 mM test solutions were used to measure open tip switching time at the end of an experiment.
To withdraw BC somas, whole cell access was obtained with a pipette filled with BC intracellular solution containing sulforhodamine 101 and Neurobiotin Tracer. Continuous negative pressure was applied to the pipette by syringe and the soma was slowly withdrawn from the slice leaving behind a soma-less but otherwise intact cell remnant that could be identified under epifluorescence or, after fixation, by confocal microscopy.
For glutamate uncaging, 5 mM MNI-caged-L-glutamate (Tocris, #1490) was added to the extracellular solution and applied to the synapse by a local puffer pipette. BCs did not respond to puffer-applied caged-glutamate in the absence of the flash. Spots, ~3 µm in diameter at the level of the slice, were delivered by a Vortran Stradus Laser (405 nm, nominal power = 50 mW) via a Rapp OptoElectric Spot Illumination System that was attached to the epifluorescence port. Laser power was set using Vortran Stradus Control Software and an electronic shutter was driven by a TTL pulse from the ITC-18. The microscope objective was an Olympus LUMPlanFLN 60x/1.00 W.
The model cone terminal contained 20 invaginating transmitter release sites of which x were contacted by dendrites. The probability of response failure, pf, was calculated as a function of the number of simultaneously released cone quanta, n, for each x between 1 and 20. Every quantum had an equal probability of release at all 20 sites. For the case where a success occurs whenever a single vesicle is released at a site that is occupied by a dendrite, the probability of failure pf(n) = (1 – x/20)n. Probability curves were calculated for each x and associated with a cb2 cell data set by least squares minimization. pf(n) plots at a given x consisting of discrete points were fitted with an exponential decay curve for display (e.g., Fig. 2g).
The model cone terminal had 19 invaginating release sites arranged in a trigonal array (Fig. 3c). Contact number could be varied from 1 to 24 with individual contacts located at the centers of triangles whose vertices were release sites. A response rule was constructed and, for a given number of dendritic contacts, the relationship between the failure probability pf and the number of released cone quanta, n, was determined by running a simulation. In a typical response rule, a success occurs when any one dendritic contact receives 2 or more (or ≥3, ≥4, or ≥5) quanta in any combination from the 3 surrounding vertices. Computationally, in a trial, each bin in an array of 19 release sites is populated with m release events where m = 0, 1, 2, 3… as determined by the Poisson distribution and a random number generator. The Poisson average in a simulation was set to n/19 where n is the user assigned total number of quanta released in the trial. To determine the pf for n cone quanta, the program generated 1000 trials while only retaining those where the sum of all released quanta across bins equaled n. For example, to determine pf for the condition in which a cone terminal releases exactly 2 quanta, the sum across all retained bins must equal 2. The retained ‘release site’ trial array was then used to populate a 24-bin “dendrite” trial array using the mapping from Fig. 3c and the response rule. For example, for the ≥2 rule above, dendrite array bin 1 would be set to success (=1) if the sum of release site array bins 2, 3, and 6 is ≥2 or failure (=0) otherwise. An analogous operation is then performed for each of the 24 dendrite bins. For each retained trial, if the sum of the dendrite array is 0, then the trial is a failure, otherwise it is a success. The proportion of failed trials was then calculated to give pf for the number of cone quanta released. The simulation was repeated for n = 1… 20 to obtain the entire probability curve. The role of contact number was examined by masking the dendrite array so that the values in the first x bins were maintained while values in the remaining bins were set to 0. To model a certain percentage of success at the single quantal release level while otherwise maintaining the ≥2 response rule, 20-35% of the bins that contained 1 vesicle in the retained array were randomly tagged so as to guarantee a success when mapped to a bin in the dendrite array. To simulate multiquantal release, each of the m events in each release array bin was associated with a Ca2+ channel opening that could randomly lead to mono-vesicular release 80% of the time and di-vesicular release 20% of the time. The total release events in the bin were adjusted accordingly. For trial retention, the sum of all quanta still had to equal n. Hence, multi-quantal release does not contribute to pf when the model cone releases only a single quantum (n = 1). Fits were evaluated by least squares minimization against data points and the best fit was arrived at through an iterative procedure.
Tissue was either fixed in 4% PFA using standard procedures6 or in 2% glyoxal solution81. Glyoxal fixative (10 ml) contained 7.325 ml distilled H2O, 2 ml absolute ethanol, 0.5 ml glyoxal (40% in water, Sigma-Aldrich, #128456), 0.075 ml glacial acetic acid, and 0.1 ml Na-acetate (3 M, Ambion, #AM9740) to give a pH of ~4.081. Pieces of freshly dissected retina with pigment epithelium, 1 × 1 mm, were placed ganglion cell side down onto dry Millipore filter paper (3.0 µm MCE Membrane, #SSWP02500) in a small petri dish. Cold PBS was added, and the pigment epithelium was peeled from the paper-adherent retina. The petri dish was placed on ice for 5 min and then cold glyoxal solution was substituted for the PBS. After 2 hrs on ice, the tissue was placed in the refrigerator at 4 oC for 2 days. Tissue was removed from the refrigerator and washed 3x for 20 min each with cold DPBS (Gibco, #21600-010), and sliced with a tissue chopper into 100 µm or 300 µm thick sections for cross-sectional or whole mount views, respectively. Individual slices were transferred to MatTek 3 mm glass bottom culture dishes and incubated for 1 day in block solution containing 0.5% Triton X-100, 0.1% Na-azide, and 3% donkey serum in 0.1 mM phosphate buffer. Whole mounts were treated with primary antibodies for 6 days (4 days for slices) in block at 10 oC with gentle shaking. The tissue was then washed 6x for 30 min each and incubated in secondary antibody solution with block for 4 days (2 days for slices) with shaking at 10 oC. The tissue was then washed 6x for 30 min each with 0.1 M phosphate buffer plus 0.1% Na-azide in preparation for mounting. Tissue was mounted in ~1 g of melamine resin consisting of 0.6 g 2,4,6-Tris[bis(methoxymethyl)amino]-1,3,5-triazine (melamine, TGI America, #T2059), 80 mg citric acid monohydrate (Sigma-Aldrich, #C1909), 20 mg of 8,000 MW polyethylene glycol (2%, w/w, Sigma-Aldrich, #89510), 5 mg caffeic acid (Sigma-Aldrich, C0625), 5 mg propyl gallate (Sigma-Aldrich, #02370), and 0.3 ml distilled water. The mixture was liquified by vortexing and then incubating on a horizontal shaker (200 rpm) in an oven at 55oC for 1 hr. For mounting, the tissue was rapidly washed 2x with distilled water that was then completely removed. The tissue was bathed in melamine resin for 1 hr at room temperature and then removed and transferred with a fine spatula to a glass slide. 15 µl of free resin was added to the tissue followed by coverslipping (Zeiss, High Performance, 18 x 18 mm, #1½). The resin was cured at 55 oC for 2 days after which it had a refractive index of 1.52.
Primary antibodies, sources, and dilutions are listed below. Secondary antibodies (JacksonImmuno) raised in donkey against mouse (#715-005-151), rabbit (#711-005-152), goat (#703-005-155), guinea pig (#706-005-148), and chicken (#705-005-147) were reacted with NHS-ester dyes using standard procedures (http://abberior-instruments.com/wp-content/uploads/0236_20120316-labeling_protocol.pdf). NHS-ester dyes were ATTO 532 (Atto-tec, #AD 532-31), Abberior STAR 580 (Abberior GmbH, ST580-0002), Abberior STAR 635 P (ST635P-0002), and CF680R (Biotium). Degree of labeling and final antibody concentration were determined with a spectrophotometer using published values for maximal absorption wavelengths, extinction coefficients, and correction factors at 280 nm. Secondary antibody dilution was 1:200. Primary antibodies, species, sources, and dilutions include: guinea pig anti-PSD95 PDZ domain, SySy 124014, (dil. 1:1000); rabbit anti-SLC1A7/EAAT5, Sigma, HPA049124, (1:100); goat anti-CtBP2 (C-16), Santa Cruz Biotechnology, sc-5967 (1:100); rabbit anti-GluR4, EMD Millipore, AB1508 (1:200); mouse anti-GluR5 (E-12), Santa Cruz Biotechnology, sc-393420 (1:500); goat anti-ChAT, Chemicon, #AB144P (1:100); chicken anti-GFP, Abcam, ab13970, (1:1000); chicken anti-bassoon, SySy 141016 (1:500); and rabbit anti-Alexa Fluor 405/Cascade Blue, A-5760 (1:500).
Intravitreal injection used a capsid modified AAV2 containing a single stranded DNA which coded for green fluorescent protein (GFP) under the control of a chicken β-actin promoter6. Anesthesia was produced by injecting animals with ketamine (100 mg/kg) and xylazine (10 mg/kg) IP. The target eye received one drop of dilute betadine solution which was immediately followed by one drop of proparacaine (1%). A 27-gauge syringe needle was used to make a hole at a location 1–2 mm below the limbal margin near the medial epicanthus. Virus (10–20 µl; titer equal to 1012–1013/ml) was injected by inserting a blunt 30-gauge syringe-mounted needle through the access hole into the vitreal chamber. After injection, antibiotic drops were then applied to the treated eye, and the animal was injected with meloxicam (1.0 mg/kg, SC), yohimbine (0.5–1.0 mg/kg, IM), and 0.9% NaCl solution (5–10 ml, SC). The typical incubation period was 3 weeks after which the retinas were removed and processed using glyoxal fixation.
Super resolution microscopy was performed on a Leica SP8 3D-STED system using a 100X NA 1.4 oil objective, white light laser for excitation, a 775 nm depletion laser, and z vortex set for isotropic voxel generation under control of LASX software. 3D capture used 20 nm XY and 50 nm Z steps using 16 kHz scan rate, 8 or 16 line average and varied frame accumulation to build up the signal. Image data were denoised using FIJI CSBDeep noise2void trained on each day’s data (https://imagej.net/plugins/n2v). Denoised images were deconvolved in Imaris 9.8.0 using the ClearView module. Deconvolution parameters were: Robust (iterative), 2.0 pre-sharpening gain, 10 iterations, denoising filter equal to 0.7. Specimen and medium refractive index were both set to 1.52. Unless otherwise noted, images are maximum intensity projections.
The analysis procedure used in Fig. 3f had four steps (Supplementary Fig. 6). In step 1, square regions centered around each terminal’s receptor labeling were excised from a binarized (Fiji, default settings) STED image. The size of the square region was 4 μm × 4 μm. Excised images were superimposed and averaged to obtain the spatial profile of GluK1 or labeled BC dendrites beneath the terminal. Averages were converted to 8 bit. Step 2 involved feature extraction by Fourier filtering. A two-dimensional Fourier transform was applied to the average profile, the frequency components with low power were removed, and then an inverse transform was performed. The power cutoff threshold was determined manually for each image and is indicated in the figure panels. In step 3, contour borders were abstracted from the filtered profiles using a Canny edge detection algorithm and a Hough transform, the latter by assuming a circular or elliptical profile. Canny edge detection and Hough transforms were executed using Python OpenCV modules. The fourth step was to quantify the spatial bias of GluK1 labeling or dendrite distribution using moment analysis. Within the region formed by central and peripheral contours, the centroid coordinates were calculated twice, first for a uniform distribution and then for the actual intensity distribution. In both cases, the centroid coordinates (Cx, Cy) were given by the following equation:
where xi, yi are the i-th pixel coordinates of the formed region, and R(x, y) is the pixel value (e.g., the overlap fraction of GluK1). The difference vector between the coordinates obtained from the uniform profile and actual data gives the magnitude and orientation of the bias of the GluK1 or dendritic process distribution.
The glutamate receptor models for cb1a and cb2 were based on a Markov scheme82. The behavior of the receptors was represented by transitions between nine discrete states (Supplementary Fig. 9a). The transition rates were different depending on the cone BC type (Supplementary Table 3). Previously published rates were used for the cb2 receptor model18. The rate constants for the cb1a cell receptor model are new and were obtained using Particle Swarm Optimization (PSO)83. The model output was optimized against receptor responses and parameters during various test glutamate applications. These parameters include individual response rise and decay, EC50, IC50, recovery τ’s from desensitization, and the percent open probability during long exposures to glutamate (Fig. 5b-d; Supplementary Fig. 8) using a weighted sum of the absolute errors of the real and model data. The optimization process searched for the rate parameter set that minimized the weighting function. PSO was performed by using PySwarms version 1.2.0. Maximal channel open probability was estimated using non-stationary fluctuation analysis from repeated responses of cb1a receptors to an 18 mM glutamate step administered by rapid perfusion. Timing jitter between separate responses in the 20-40 µs range was removed before processing by aligning the traces according to their rising phase.
We use MCell 3.4 and CellBlender 1.0.1 to construct a model synaptic cleft (Supplementary Fig. 9d)84,85. The synaptic cleft consisted of 20 μm × 20 μm planes corresponding to the pre- and post-synaptic membranes. The planes were separated by a 16 nm cleft37,86. Glutamate molecules were released from a point in the center of the presynaptic plane with a diffusion coefficient of 400 μm2/s18. There is substantial uncertainty about the number of glutamate molecules in a vesicle in the range of 3000–800087. Therefore, instead of releasing glutamate molecules in increments of “quanta”, we varied transmitter release in increments of 1000 molecules. The presynaptic plane was impermeable to glutamate while the postsynaptic plane was 10% permeable so that lateral glutamate diffusion in the cleft is consistent with previous work18. Permeability across the postsynaptic plane is consistent with it being composed of dozens of individual dendritic contacts each separated by an extracellular space. The planar distances of transporters and receptors from the release site was increased to simulate the effect of transmitter flow into the plane through an invagination-like cylinder 200 nm deep by 50 nm wide. Adding a cylinder instead made little difference in the outcome of the simulation. An annular patch (0.02 μm2) with 100 cb2 receptors surrounded the release site at a radius of 100 nm, and a square patch (0.04 μm2) with 350 cb1a receptors was centered 600 nm away from the release site on ‘basal’ membrane. In addition, a fan-shaped patch with 3200 EAAT5 transporters was placed between those two receptor patches in a region of the model surface corresponding to basal membrane. The transporter model was based on a simple transition between glutamate bound and unbound states in the range bounded by the measured EC50 for expressed EAAT549 and the results obtained by rapid perfusion (Supplementary Fig. 2e, f). The forward and backward rates were 107 M-1s–1 and 10 or 200 s–1, respectively. In some simulations, the transporter patch was moved to the side of the invagination opposite to the receptor patch and in others it was moved in the same direction but beyond the receptor patch.
EM reagents were purchased from Electron Microscopy Sciences (EMS; Hatfield, PA, USA). Animals were euthanized as described above. Once breathing stopped, trans-cardiac prefusion of the following physiological saline began (in mM): NaCl 120, KCl 3, MgSO4 2.5, glucose 10 CaCl2 2, NaHCO3 25, equilibrated with CO2/95% O2 to a pH of 7.4, and administered at a flow rate of ~15 mL/min. Next, trans-cardiac infusion of the following aldehyde solution: 0.6% paraformaldehyde and 2% glutaraldehyde in 1X PBS, was delivered via trans-cardiac infusion for 10 min. The first 10 ml of the fixative was perfused in the first 1 min, and the remaining 10 ml were delivered over 5 min while the superior vena cava was clamped to increase perfusion pressure in the head. The eyes were removed, and the superior and inferior retina was trimmed away from the peripheral retina. Retinal pieces were immersed in the aldehyde fixative solution and incubated on ice overnight. After 24 h, retinal pieces were washed in 8% sucrose in 1X PBS at 4 °C to remove free aldehydes. The retinal pieces were then equilibrated in 2X PBS for 20 min and brought to RT. The tissue was postfixed with 2% OsO4 in 1X PBS for 60 min at RT on shaker table and washed in 1X PBS and H2O. En bloc counterstaining and dehydration were performed at RT and carried out as previously described44 with a few minor changes. Counterstaining consisted of three successive 15 min exchanges in 40 mM maleate buffer (pH 5.2) followed by a 30 min incubation in 2% uranyl acetate (w/v in maleate buffer) in the dark. Next, the tissue was rinsed in pure H2O immediately before starting the following dehydration series in EtOH: 50% EtOH 10 min, 70% for 10 min, 90% for 15 min, and 100% 10 min. Finally, the tissue was dehydrated in 100% propylene oxide (PPO) for 10 min. Infiltration into EPON812 was carried out by placing samples on a tissue rotor followed by a series of infiltration steps: (1) 1:1 PPO:EPON812, 3 hrs RT; (2) 70% EPON812 in PPO for over 5 hrs at RT; (3) 100% EPON812 for 3 h RT. Infiltrated retina was further dissected such that it could be placed flat-mounted in molds filled with freshly prepared 100% EPON812. Samples were placed in a 40 °C oven and cured overnight. The temperature was then increased to 65 °C for at least 12 h. Only portions of superior retina were further sectioned and analyzed, which is the same region of the retina that we recorded from with electrophysiological physiological methods. Ultrathin sections were made as described previously and counter stained in lead citrate44. Sections were made vertically and tangentially through the retina: perpendicular to the OPL, and tangential to the OPL. Transmission electron microscopy was performed with an FEI Tecnai Spirit G2 (Center for Advance Microscopy, Northwestern University School of Medicine). Analysis of EM images was performed using ImageJ software (NIH, Bethesda, MD, USA).
Additional statistics for Fig. 10b.
Above, for cb1b and cb2 current responses (the average of 3 complete stimulus sequence repeats) during all decrements, the hypothesis that the levels are the same was rejected (cb1b: critical value = 1.96, right to left t value = 31.27, 23.52, 53.19, 82.02, 111.63; cb2 t value = 23.55, 21.38, 57.74, 86.27, 124.88; n = 2000 points (i.e., 400 ms) before versus 2000 during the stimulus; two-tailed unpaired t test). Cb1b and cb2 normalized current responses were different from each other (critical value = 1.65, right to left t value = 3.00, 10.66, 35.97, 31.87, 67,38; two-tailed unpaired t test).
Upper middle, Levene’s Mean Statistic (LMS) tests the hypothesis, H0, that the variance before and during a light decrement are equal (α = 0.05, critical value = 3.844; cb1b, right to left, LMS = 6.54 H0 rejected but Δvar was -0.12 pA2, 3.484 H0 accepted, 53.186, 590.245, 1067.06; cb2 LMS = 115.14, 306.51, 407.96, 988.35, 978.58).
Lower middle, unitary response amplitude was calculated separately for each of the three stimulus sequence repeats and plotted at each step-to intensity. Mean±S.E. was also plotted. A two-tailed one sample t test was used to determine whether the size of the cb1b cell unit was different from 0 (cb1b: right to left, p=not calc., 0.8922, 0.0738, 0.1025, 0.0005). When the 6 unitary response amplitudes during decrements 4 and 5 were grouped together, the mean was different from 0 (p = 0.0017). During the smallest decrement, the mean cb2 cell unit was not significantly different either from 0 (p = 0.1071) or from the mean value of the other 4 decrement responses (7.9 pA, p = 0.4849; two-tailed one sample t test).
Below, mean cb1a/b cell unit amplitude different from decrement 5 (right to left, p = 0.0178, 0.0045, 0.0052, 0.3048; two-tailed unpaired t test).
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
