Autism-associated neuroligin-3 mutations commonly disrupt tonic endocannabinoid signaling.

The R451C KI enhances GABAergic synaptic transmission at CCK basket cell synapses

We next analyzed the properties of transmission at pyramidal synapses formed by CCK basket cells. Surprisingly, here the R451C KI caused a ~100% increase in the IPSC amplitudes and a ~15% increase in the IPSC success rate during 1 Hz stimulation, and a slightly smaller change during 2 and 10 Hz stimulation (Figs. 2A and 2B). The increase in success rate suggests an increase in the presynaptic GABA release probability, which is also a plausible explanation for the increase in IPSC amplitudes. This hypothesis was further supported by the absence of detectable changes in the IPSC half-width, indicating that the GABA-receptor subunit composition or uptake mechanisms were unaltered (Fig. 2C; WT: 6.3±0.4 ms, R451C: 5.4±0.3 ms). Furthermore, the amplitude of minimal unitary IPSCs (Fig. 2D; WT: 23.4±4.3 pA, R451C: 29.6±4 pA) and the rate of finding connected pairs (Fig. 2E; WT: 2.6±0.7, R451C: 2.2±0.3) were similar in wild-type and R451C mutant slices, as was the morphology of their CCK basket cells (Fig. 2F). The phenotype of the R451C mutation in the CCK cell synapses again was more consistent with a presynaptic change (such as increased release probability) than a structural alteration (e.g. increase in synapse density) or postsynaptic effect. Thus, the R451C KI produces opposite changes at two different perisomatic inhibitory synapses, and in both cases the changes appear to involve an ultimately presynaptic mechanism, even though NL3 is a postsynaptic molecule.

Figure 2

Neuroligin-3 R451C substitution enhances GABAergic synaptic transmission in CCK basket cell synapses

A synaptic phenotype of NL3 KO mice

To test whether the R451C KI phenotypes represent gain- or loss-of-function effects, we next performed paired recordings in acute slices from NL3 KO mice, again using littermate wild-type mice as controls. When we analyzed the properties of transmission between PV basket cells and pyramidal neurons, we failed to detect a phenotype. Specifically, the amplitude and success rate of IPSCs were unchanged (Figs. 3A and 3B), as were the half-width of the IPSCs (Fig. 3C; WT: 4.7 ± 0.2 ms, NL3 KO: 5.5 ± 0.4 ms), the size of unitary minimal IPSCs (Fig. 3D; WT: 17.6±1.6 pA, NL3 KO: 16.5 ± 1.1 pA), and the rate of finding connected pairs (Fig. 3E; WT: 2.2 ± 0.3, NL3 KO: 2.5 ± 0.6). These results suggest that the loss of synaptic transmission at this synapse in R451C mutant mice represents an active suppression of synaptic transmission by a gain-of-function activity of R451C-mutant NL3.

Figure 3

Neuroligin-3 KO does not alter GABAergic transmission in PV basket cell synapses

We then examined the effect of the NL3 KO on synaptic transmission mediated by inhibitory synapses that were formed by CCK-containing terminals on pyramidal neurons (Fig. 4). Surprisingly, here the NL3 KO phenocopied the R451C KI. Specifically, the NL3 KO caused a significant increase in synaptic strength, as manifested by both an increase in IPSC amplitude and in success rate (Figs. 4A and 4B). In addition, we observed a small increase in IPSC half-width (Fig. 4C; WT: 4.9±0.1 ms, NL3 KO: 5.6±0.2), but no change in the size of unitary minimal IPSCs (Fig. 4D; WT: 25.3±1.7 pA, NL3 KO: 29±2.9 pA), or in the rate of finding synaptically connected pairs of neurons (Fig. 4E; WT: 2±0.2, NL3 KO: 1.8±0.2). The fact that increased synaptic transmission at CCK basket cell synapses is equally observed in NL3 KO and R451C KI neurons shows that it is caused by a loss-of-function mechanism.

Figure 4

Neuroligin-3 KO enhances GABAergic synaptic transmission in CCK basket cell synapses similar to the R451C KI

NL3 R451C KI lowers the probability of GABA release at PV basket cell synapses

The change in success rates in our paired recordings of synapses with the NL3 R451C KI or the NL3 KO mutations suggests a presynaptic origin for the observed phenotypes, despite the postsynaptic localization of NL3 (Budreck and Scheiffele, 2007). To evaluate whether presynaptic changes alone (such as in the probability of release) could in principle account for the NL3 related phenotypes, we analyzed these phenotypes by modeling and computer simulations.

We first plotted bin-averaged PV basket cell IPSC amplitudes against their corresponding averaged success rates (Fig. 5A; inset shows distribution of individual pairs). We then fitted these data with an equation that relates IPSC amplitudes to the success rate of transmission ([PSC=Q⋅N⋅[1−1−successes‾‾‾‾‾‾‾‾‾‾‾‾‾√N]see Experimental Procedures and Figs. S2A–S2C) to estimate characteristic mean quantal size (Q) and number of release sites (N) for WT (n=25) and R451C KI (n=26) populations. The resulting estimates for Q were similar for both genotypes (mean and 95% confidence intervals: 21.9/12.5–31.3 pA and 17.5/16.4–18.6 pA for WT and R451C, respectively), as were the estimates for N (mean and 95% confidence intervals: 7.6/1.6–13.6 and 8.7/0–17.4 for WT and R451C synapses, respectively). These estimates support the notion that the synaptic phenotype in R451C-mutant PV basket cell synapses was not due to a decrease in quantal size (see Fig. 1D), and, limited by the wide confidence interval of estimates, also suggest that the R451C phenotype was not due to a decrease in the number of release sites.

Figure 5

The NL3 R451C KI mutation lowers the probability of GABA release from PV basket cell synapses

To examine the remaining possibility, namely that a lower neurotransmitter release probability (P_R) underlies the R451C phenotype, we performed computer simulations in which we modeled IPSCs at different P_R values (Figs. 5B and S3). In this computational model, we incorporated a minimal set of synaptic parameters that allowed us to simulate the IPSC amplitudes and success rates, and to compare these parameters to the experimental data. The simulation parameters included, in addition to the number of release sites (N), the mean and the variance of the release probability (P_R and σ_PR) and the mean and the variance of the quantal amplitude (Q and σ_Q). For each simulated paired recording, the computationally determined IPSC (cIPSC) was derived as cIPC=∑Ni=1pi⋅qi, and the computationally determined success rate (cSuccesses) was derived as cSuccess=[1−∏Ni=1(1−pi)]⋅100, where p_i and q_iare the probability of release and the quantal amplitude in the i-th release site, respectively (see Experimental Procedures and Figs. S3A–3G).

We started the simulations by using Q and N values estimated from the population quantal analysis (Fig. 5A; see above) to derive values for P_R, σ_PR, and σ_Q that result in cIPSCs and cSuccesses which approximate the experimentally determined IPSCs and success rates. For PV basket cell IPSCs in WT neurons, we found that a P_R=0.23, together with a σ_PR=0.224 and a σ_Q=2.25 (Q=21 pA and N=7, per modeling), provided computationally determined cIPSCs and cSuccesses that did not significantly differ from the experimental data (mean difference ± SD for IPSCs: 0±6 pA; for success rates: 0±0.02; t-test, P>0.5 for both). For computer simulation of R451C synapses, we found that much lowered release probabilities, P_R=0.11, together with a σ_PR=0.09 and a σ_Q=1.65 (Q=17 pA and N=8, per modeling) were needed to replicate the experimental data (mean difference ± SD for IPSCs: 0±0.99 pA; for success rates: 0±0.01; t-test, P>0.5 for both). These simulations thus suggest that a ~2-fold decrease in the probability of GABA release could sufficiently explain the NL3 R451C KI phenotype in PV basket cell synapses. These conclusions were further supported by consequent analysis of biocytin-filled axons (Fig. 5C), which also did not indicate a difference in the number of synapses formed by individual PV basket cells (WT: 0.33±0.03 and R451C: 0.26±0.01, synapses per μm, Mann-Whitney RST, P=0.152).

Next, we sought to determine a cause for lower release rates in PV basket cell synapses in the R451C KI mice. We reasoned that such decreases in release rate could be caused by NL3 mutation-driven alterations of the presynaptic release machinery, or alternatively, by over-activation of a presynaptic receptor, such as a neuropeptide receptor, that physiologically suppresses GABA release from these synapses (Freund and Katona, 2007). We addressed this latter possibility by application of pharmacological agents in paired recording experiments.

Activation of two presynaptic G-protein coupled receptors, namely μ-opioid and M2 muscarinic-receptors, is known to suppress GABA release at PV basket cell synapses (Glickfeld et al., 2008, Szabó et al., 2010). Thus, we tested the effect of the μ-opioid receptor antagonist CTAP (500 nM; Fig. 5D, n=4 pairs) and of the M2 muscarinic-receptor antagonist AF-DX (10 μM; Fig. 5E, n=4 pairs) in paired-recordings of PV basket cell to pyramidal neuron synapses in NL3 R451C KI mice. Neither antagonist increased IPSC amplitudes in paired recordings, indicating that tonic activation of these receptors does not account for the decreased transmission at PV basket cell synapses in R451C KI mice. In additional control experiments, both antagonists reliably reversed the effect of their corresponding agonists, DAMGO (1 μM) and carbachol (5 μM; not shown). Thus, the presence of NL3 the R451C mutation likely induces a functional change in the presynaptic release properties of PV basket cell synapses.

Neuroligin-3 is essential for tonic endocannabinoid signaling at CCK basket cell synapses

Our data suggest that a loss of NL3 function produces an increase in GABA release at synapses formed by CCK basket cells onto pyramidal neurons synapses. CCK basket cell synapses exhibit a distinct feature that offers an immediate hypothesis to account for the observed phenotype. This feature consists of the efficient suppression of GABA release from CCK basket cell terminals by the endocannabinoid-mediated activation of presynaptic CB1 receptors (reviewed in Alger, 2002; Piomelli, 2003; Freund and Katona, 2007).

Endocannabinoids are secreted from postsynaptic pyramidal neurons to activate presynaptic CB1 receptors in two principal modes. Phasic secretion of endocannabinoids is induced by postsynaptic depolarization and/or mGluR5 activation and mediates decreases in synaptic transmission during short- and long-term plasticity. Tonic secretion of endocannabinoids affects synaptic transmission over longer time periods (reviewed in Alger, 2012; Katona and Freund, 2012). A deficiency in tonic endocannabinoid signaling, with or without an effect on phasic endocannabinoid signaling, would be expected to enhance the probability of GABA release, and thus would increase IPSCs similar to what we observed in R451C KI and NL3 KO neurons. Thus, we tested the hypothesis that a loss-of-function of NL3 – either via the KO or via the R451C KI – impairs tonic endocannabinoid signaling.

In wild-type synapses, bath application of 10 μM AM251 (a CB1 receptor antagonist and inverse agonist) caused a ~100% increase in IPSC amplitudes and ~50% increase in success rate (Figs. 6A and 6B; 1 Hz AP firing), reflecting disinhibition of GABA release by blocking tonically active CB1 receptors (Neu et al., 2007). In NL3 KO synapses, strikingly, AM251 did not enhance IPSC amplitudes (Figs. 6A, S4A, and S4B) or success rates of synaptic transmission (Figs. 6B, S4A, and S4B). These findings suggest that IPSC amplitudes in the NL3 KO were larger because these synapses express higher release probabilities due to an apparent lack of tonic CB1 receptor activation.

Figure 6

Neuroligin-3 KO and R451C KI mutations impair tonic endocannabinoid signaling

To evaluate whether differences in the release probability alone, without other possible consequences of NL3 deletion, could explain the observed phenotype, we again used modeling and computer simulations. Fitting of the bin-averaged IPSC – successes data (Figs. 6C and S3A–S3C) resulted in similar Q and N estimates for the NL3 WT and KO data sets (mean and 95% confidence intervals; Q: 39 / 30.8–47.3 and 46.2 / 14.1–78.4 pA, and N: 5.6 / 4.1–7 and 4.4 / −2.9–11.8, for WT and NL3 KO, respectively). Using these parameter estimates in subsequent simulations (Figs. 6D and S3A–S3G), we found that the mean values of simulated IPSC – successes distributions were not significantly different from experimental values (inset in left panel) when P_R=0.12 (together with a σ_PR=0.19 and a σ_Q=2; Q=39 pA and N=6 per model estimates) for NL3 WT, and when P_R=0.26 (together with a σ_PR=0.26 and a σ_Q=2.1; Q=46.2 pA and N=5 per model estimates) for NL3 KO. In addition, we quantified axonal bouton densities (Fig. 6E), which were not different between the two genotypes (WT: 0.18 ± 0.01 and NL3 KO: 0.18 ± 0.01, per μm, t-test, P=0.779). Together, these analyses suggest that the loss of tonic CB1 receptor activation, and the consequent ~2-fold increase in the probability of GABA release, is sufficient to account for the entire phenotype of the NL3 deletion at these synapses.

We next determined whether the loss of tonic CB1 receptor activation was affecting GABA release only from basket cell synapses, or whether all CB1-containing GABAergic synapses exhibit this phenotype. Thus, we repeated the CB1 receptor blocking experiments by monitoring IPSCs evoked by extracellular stimulation (which will cause GABA release from a broad set of presynaptic fibers that include CB1-receptor-containing axons). Application of AM251 again enhanced IPSCs in CA1 pyramidal cells, but consistently failed to do so in the NL3 KO (Fig. 6F). We also repeated these latter extracellular stimulation experiments with CP 945,598, a CB1 receptor antagonist that is structurally unrelated to AM251. Bath application of CP 945,598 (5 μM) replicated the findings with AM251 (Fig. 6G), independently confirming the absence of tonic EC signaling in NL3 KO mice.

Similar to the NL3 KO, paired recordings from slices prepared from the NL3 R451C KI mice revealed that the effect of AM251 on CCK basket cell IPSCs was greatly reduced (Figs. 6H and 6I). These data suggest that NL3 is essential for the tonic endocannabinoid signaling that inhibits GABA release from CCK basket cell synapses. Futhermore, we tested whether the NL3 KO may alter tonic CB1 receptor-mediated signaling at glutamatergic synapses. We stimulated Schaffer-collateral synapses and recorded from CA1 pyramidal cells (in the presence of 50 μM picrotoxin). However, bath application of AM251 (10 μM) failed to increase EPSC amplitudes in either WT slices or NL3 KO slices (Fig. 6J; see also Hoffmann et al., 2010). Together, these data suggest that NL3-related mutations may impair tonic endocannabinoid signaling at CB1 receptor-containing inhibitory, but not excitatory synapses.

Electrophysiology

Hippocampal slices (300 μm) were prepared from 3–4 weeks old NL3 R451C KI and NL3 KO mice. Slices were incubated at 33 °C in sucrose-containing artificial cerebrospinal fluid (ACSF; 85 mM NaCl, 75 mM sucrose, 2.5 mM KCl, 25 mM glucose, 1.25 mM NaH₂PO₄, 4 mM MgCl₂, 0.5 mM CaCl₂ and 24 mM NaHCO₃) for an hour and than incubated in the same solution at room temperature until recording. Electrophysiological recordings were made in ACSF containing 126 mM NaCl, 2.5 mM KCl, 10 mM glucose, 1.25 mM NaH₂PO₄, 2 mM MgCl₂, 2 mM CaCl₂ and 26 mM NaHCO₃. Slices were visualized in an upright microscope (Olympus, BX-61WI) with infrared differential interference contrast optics. Whole cell recordings were obtained from the interneurons with patch pipettes (King Precision Glass, Inc., 3–5 MΩ) filled with internal solution containing 126 mM K-gluconate, 4 mM KCl, 10 mM HEPES, 4 mM Mg-ATP, 0.3 Na-GTP, 10 mM phosphocreatine and 0.2% biocytin (pH 7.2, 270–290 mOsm), and from postsynaptic pyramidal cells containing 40 mM CsCl, 90 mM K-gluconate, 1.8 mM NaCl, 1.7 mM MgCl₂, 3.5 mM KCl, 0.05 mM EGTA, 10 mM HEPES, 2 mM Mg-ATP, 0.4 mM Na-GTP, 10 mM phosphocreatine (pH 7.2, 270–290 mOsm; in some of the recordings 0.2% biocytin was also added to this solution). All electrophysiological recordings were made at 33 °C, using MultiClamp700B amplifiers (Molecular Devices, Sunnyvale, CA). Signals were filtered at 4 kHz using Bessel filter and digitized at 10 kHz with a Digidata 1440A analog-digital interface (Molecular Devices, Sunnyvale, CA). Series resistance was monitored, and recordings were discarded if the series resistance changed significantly or reached 25 MΩ. The recorded traces were analyzed using Clampfit software (Molecular Devices, Sunnyvale, CA). PV and CCK interneurons were distinguished based on their distinct electrophysiological spiking properties (Földy et al., 2010), and by the presence of DSI in CCK basket cell synapses (see Fig. 7A). IPSCs were individually inspected and included in the analysis based on their onset latency following the presynaptic action potential. For statistical analysis Student’s t-test, paired t-test or Mann-Whitney Rank Sum Test (RST) was used, and data are presented as mean ± s.e.m., unless noted otherwise; significance was P < 0.05.

Quantal model

Individual basket cells innervate postsynaptic pyramidal cells via multiple release sites (N; Biró et al., 2006; Földy et al., 2010), in which intrinsically variable synaptic parameters (such as quantal size and release probability; Q and P_R respectively) produce a trial-to-trial fluctuation in the IPSC amplitudes. The distribution of these fluctuations can be described by models that are based on binomial statistics and allow estimates of Q and N (Silver, 2003, Biró et al., 2006). In this study, we ought to extend quantal modeling to analyze pooled data from multiple paired-recording experiments of defined synapse populations, and extract mean quantal information that is characteristic to each population. For modeling, we analyzed synapses by quantifying IPSC amplitudes and success rates. Assuming that each synapse population can be described by characteristic mean N and Q values, it is reasonable to assume that the pair-to-pair variability in IPSC amplitudes and success rates is dominated by variability in P_R. In this case, the distribution of IPSC amplitudes and success rates should follow the [PSC=Q⋅N⋅[1−1−Successes‾‾‾‾‾‾‾‾‾‾‾‾‾√N]model (Eq. 1; see Fig. S2 for more information). For fitting the IPSC model on experimental data, to estimate quantal parameters, we employed the built-in, unconstrained NonlinearModelFit algorithm in Mathematica 8 (Wolfram Research, Inc., Champaign, IL). Note that the basic assumptions of this approach (i.e. the existence of characteristic Q and N values in each synapse population) were supported by the similarity between the observed and predicted IPSC distributions (Eq.1; see Figs. 5A and ​and6C6C).

Computational model

In order to gain further qualitative insight into how pre- and postsynaptic changes may contribute to the synaptic phenotypes produced by NL3 mutations, we devised a simple computational model that incorporated five modifiable synaptic parameters: the number of release sites (N) and the mean and variance of the release probability (P_R and σ_PR, respectively) and of quantal IPSCs (Q and σ_Q, respectively). Note that non-zero variances were necessary to simulate variability both in the number of successful transmissions (by σ_PR) and IPSC amplitudes (by σ_Q). To initialize the simulations, p_i values (that is the release probability of the i-th release site) were assigned randomly from a normal probability distribution function with P_R mean and σ_PR variance for each release site. In addition, for each release site, q_i values (that is the quantal size in the i-th release site) were randomly assigned from a log-normal probability distribution function of mean Q and σ_Q variance parameters (see Supplementary Fig. S3 for more information). Computational IPSCs (cIPSCs) and successes (cSuccesses) were derived as described in the text. For each condition, estimates of Q and N were adopted from the quantal model (Fig. 5A and ​and6C).6C). Each simulation had the same sample size as the original data, and each simulation was repeated 50 times with random assignments of new p_i and q_i values. For statistical comparisons, we tested the null-hypothesis that the difference between the mean computed and experimental successes and IPSCs were zero; simulation parameters were accepted when P>0.05 using Student’s t-test. To estimate the robustness of the resulting simulation parameters, we quantified an average range for each parameter which still justifies the null-hypothesis: ΔP_R=±0.006, Δσ_PR=±0.09, ΔQ=±0.7 pA and Δσ_Q=±0.06 pA (relative to values presented in the main text). Parameter deviations beyond these ranges independently resulted in statistically significant differences (P<<0.05) between the simulated and experimental distributions. Simulations were implemented and run using Mathematica 8 (Wolfram Research, Inc.).

We would like to thank Nathan Huang, Scarlett Fang, Ayeh Darvishzadeh and Shaon Ghosh for technical assistance, and Dr. Jason Aoto for advice. This study was supported by grants from the Simons Foundation (177850 to TCS), the NIMH (P50 MH086403 to RCM and TCS), the NIDA (K99DA034029 to CF), and the NINDS (NS069375 to the Stanford Neuroscience Microscopy Service).