Reduced expression of the psychiatric risk gene DLG2 (PSD93) impairs hippocampal synaptic integration and plasticity

Copy number variants indicating loss of function in the DLG2 gene have been associated with markedly increased risk for schizophrenia, autism spectrum disorder, and intellectual disability. DLG2 encodes the postsynaptic scaffolding protein DLG2 (PSD93) that interacts with NMDA receptors, potassium channels, and cytoskeletal regulators but the net impact of these interactions on synaptic plasticity, likely underpinning cognitive impairments associated with these conditions, remains unclear. Here, hippocampal CA1 neuronal excitability and synaptic function were investigated in a novel clinically relevant heterozygous Dlg2+/− rat model using ex vivo patch-clamp electrophysiology, pharmacology, and computational modelling. Dlg2+/− rats had reduced supra-linear dendritic integration of synaptic inputs resulting in impaired associative long-term potentiation. This impairment was not caused by a change in synaptic input since NMDA receptor-mediated synaptic currents were, conversely, increased and AMPA receptor-mediated currents were unaffected. Instead, the impairment in associative long-term potentiation resulted from an increase in potassium channel function leading to a decrease in input resistance, which reduced supra-linear dendritic integration. Enhancement of dendritic excitability by blockade of potassium channels or activation of muscarinic M1 receptors with selective allosteric agonist 77-LH-28-1 reduced the threshold for dendritic integration and 77-LH-28-1 rescued the associative long-term potentiation impairment in the Dlg2+/− rats. These findings demonstrate a biological phenotype that can be reversed by compound classes used clinically, such as muscarinic M1 receptor agonists, and is therefore a potential target for therapeutic intervention.

Methods: brain slice preparation, electrophysiology, protein quantification and computational modelling    Prior to the generation of the founder rats an initial in-vitro confirmation of the efficiency of the sgRNA-Cas9 was demonstrated by nucleofecting the sgRNA-Cas9 into rat C6 glial cells.
Genomic DNA (gDNA) PCR products were subsequently generated from nucleofected C6 cells using primers flanking the sgRNA site. gDNA PCR products were screened for nonhomologous end joining activity and deletion mutations using the SURVEYOR Cel-1 Mutation Detection Assay (Integrated DNA Technologies). Founder rats were then produced as follows; embryo donor female Long Evans rats were super-ovulated with pregnant mare serum (PMS) and given human chorionic gonadotrophin (HCG) 48 hrs post PMS administration. Females were immediately mated to stud males after HCG administration. Embryo donor females were euthanized 18-24 hrs after mating and their one-cell fertilized embryos were isolated by harvesting the reproductive tract and rupturing the ampulae. Harvested embryos were put in culture media in a CO2 incubator until ready for microinjection. One-cell stage embryos were microinjected with the validated sgRNA-Cas9 and then implanted into synchronized pseudopregnant Long Evans recipient females.
The success of the targeting strategy was confirmed in founder rats using sequencing of gDNA PCR products derived from P14 tissue biopsies. Fig S2 illustrates genomic sequencing of a Dlg2 7bp out of frame heterozygous deletion founder. Manual reading of each double peak in the sequencing chromatograph (ABI Sequence Scanner) shows that upstream of the deletion, sequences from the wild-type and modified allele are identical. At the site of the deletion i) the sequence read becomes mixed and ii) the sequence of the secondary peaks from the modified allele align with wild-type sequence, except they occur 7bp further upstream revealing the size and position of the modified allele. As detailed in Fig S3-  Possible off-target hits were assessed by generating a list of top 10 potential off-target (OT) sites, based on the sgRNA sequence used (CCAGGGTCATCTCCAATGTGagg) and ranked using the MIT website http://crispr.mit.edu/. The top 10 OT sites were computed by taking into account the following i) total number of mismatches, ii) mismatch absolute position (to accommodate for the relatively high disturbance of mismatches falling close to the PAM site) and iii) mean pairwise distance between mismatches to account for the steric effect of closely neighbouring mismatches in disrupting sgRNA-DNA interaction. Corresponding PCR primer pairs were designed to flank the top 10 potential off target sites. Using extracted gDNA from the founder animal and wild-type controls, gDNA products were generated to flank each potential OT site (~300-500bp amplicon) and run on the SURVEYOR assay. Confirming the specificity of the CRISPR-Cas9 targeting to exon 5 of the Dlg2 gene none of the 10 OT sites tested revealed NHEJ activity (Table S1).
Founder rats were mated with wild types at Horizon Discovery and generated F2 progeny containing the mutation, thus confirming germ-line transmission. A total of five male positives were exported to Charles River, Lyon, France for re-derivation by embryo transfer. The resulting specific pathogen free (SPF) progeny were sent to Charles River, Margate, UK for routine breeding and maintenance of the lines. The standard breeding protocol was a heterozygous x wild-type cross giving rise to 1:1 average Dlg2+/-/wild-type progeny allowing full use of the litter and the generation of littermate controls. The Dlg2+/-rat model is viable and Charles River has reported no adverse effects on breeding performance, development, general health and in addition, no deviation from the expected Medelian 1:1 ratio of Dlg2+/-to wild-types and no skewing of the sex ratios. For the current experiments a cohort of breeders was transported to University of Bristol Animal Service Unit and the experimental animals used in the work were generated using heterozygous x wild-type cross giving rise to 1:1 average Dlg2+/-/wild-type littermate control progeny.

Whole-cell patch-clamp recordings
The recording chamber was perfused with oxygenated aCSF at 32 °C. Slices were visualised using a differential interference contrast Scientifica SliceScope microscope. Borosilicate glass pipettes (pipette resistance of 4-7 MΩ) were pulled using a horizontal P-97 Sutter-instruments Recordings were obtained using a Multiclamp 700A Molecular Devices amplifier, filtered at 2.4 kHz, and sampled at 10, 20, or 25 kHz, depending on experiment, using a Cambridge Electronic Designs Micro 1401 data acquisition board. Cambridge Electronic Designs Signal 5.12 and Spike 2.7 software were used for data acquisition. All data are presented without adjustment for the junction potential (~−15 mV).

Synaptic stimulation and plasticity protocols
All recordings were made in the presence of picrotoxin 50 µM. Schaffer collateral (SC) and temporoammonic (TA) fibres were alternatively stimulated using a paired pulse protocol.

Experiment: AMPA:NMDA ratio
Using the caesium-based internal solution, cells were held at -70 mV for 10 min, +40 mV for 5 min, and -70 mV for another 10 min. This allowed for the acquisition of predominantly AMPAmediated EPSCs at -70 mV and combined AMPA-and NMDA-mediated EPSCs at +40 mV.
AMPA-mediated EPSCs were measured at the peak amplitude, whilst NMDA-mediated EPSCs were measured at 45 ms after the first stimulation artefact. If the AMPA-mediated EPSCs recorded before and after the voltage switch differed by more than 30%, data was excluded from the analysis.
Experiment: AMPA mEPSC mEPSCs were recorded using the caesium-based internal solution. Recordings were made in the presence of TTX 500 nM and at a holding potential of -65 mV. 10 µM NBQX was applied in a subset of the initial experiments to confirm that the mEPSCs were AMPA-mediated ( Fig   S5). The following parameters were used for mEPSC detection post hoc: -3 pA amplitude, 0.1 ms tau (rise), 3 ms tau (decay), dead time 7 ms, rising edge window 2 ms.

Experiment: SK-mediated modulation of NMDAR
With a potassium-based QX-314 internal solution, cells were held in current-clamp with enough current injection to keep them at approximately -50 mV. The aCSF contained CGP55845 1µM throughout the experiment, with subsequent applications of apamin 100 nM alone and together with and D-APV 50 µM for 10 min each. Compound EPSPs were elicited using a burst of 5 stimulations at 100 Hz. EPSP decay was modelled using a single exponential function fit between the linear component of the decay of the final EPSP and baseline, from the averaged traces of the final 3 min of every condition.

Experiment: GluN2b-mediated EPSC decay
With a caesium-based internal solution, cells were held at +40 mV. aCSF contained NBQX 10 µM throughout the experiment, with subsequent additions of RO256981 1 µM alone and in combination with D-APV 50 µM for 10 min each. The decay of the second EPSC was modelled using a double exponential function fit between the amplitude peak and the baseline, from the averaged traces of the final 3 min of every condition.

Experiment: Paired theta burst LTP in SC synapses
Using a potassium-based internal solution, SC and TA fibres were alternatively stimulated in the voltage clamp configuration at -65 mV. With a 5 min baseline and within 10 min of breaking into the whole-cell configuration, LTP was induced in the SC synapses using a 2 second paired theta burst protocol (5 Hz bursts of 5 stimuli at 100 Hz) with simultaneous presynaptic stimulation and somatic depolarization (2 ms, 1 nA) steps. LTP induction was performed in the current-clamp configuration with enough current injection to maintain the cell at -65 mV.
LTP was measured at 25-30 min post induction, using the TA pathway as a negative control.

Experiment: Associative LTP in SC and TA synapses
Using a potassium-based internal solution, the cell was held at -65 mV. SC (from either side of the recording cell) and TA fibres were alternatively stimulated. Prior to aLTP induction, stimulation strength was adjusted to produce EPSCs with a mean amplitude of ~100 pA. With a 5 min baseline and within 10 min of breaking into the whole-cell configuration, LTP was induced in the one of the SC and the TA synapses using a 2 s theta burst (5 Hz) protocol with simultaneous presynaptic stimulation of the SC and the TA fibres, with no somatic depolarization steps. The SC pathway from the other side of the recording cell (the one that did not participate in LTP induction) was used as a control pathway. The cell was maintained at -55 mV in current clamp during induction and the theta burst train was repeated for a total of 3 times, at a 20 s interval. Initial baseline EPSC amplitudes were adjusted to be approximately 100 pA. LTP was measured at 25-30 min post induction. A pathway check was run at the end of the experiment to ensure the synapses stimulated belonged to distinct pathways ( Fig S6). The experiment was repeated in the presence of 77-LH-28-1 7 µM, with baseline EPSC amplitudes adjusted to be approximately 100 pA. In the induction trace analysis, spikes were identified by finding local maxima over -20 mV. 2 ms before and 4 ms after the identified peaks, segments of the traces were removed. The missing values were interpolated to produce an approximation of the underlying EPSP.

Experiment: Supralinear integration of dendritic spiking into plateau potentials
Using a potassium-based internal solution containing QX-314, a single and a compound EPSP (burst of 5 stimulations at 100 Hz) in the SC pathway were recorded using increasing stimulus intensity in the presence of CGP55845 1µM. The subthreshold rising EPSP slope of the single EPSP was used to normalise area under the curve comparisons of the compound EPSP across conditions. The area under the curve was calculated from 50 ms to 450 ms after the compound EPSP. The slope-area relationship was smoothed using Savitzky-Golay filtering.
The data was then passed through a change point analysis function in Matlab designed to find changes in signal to identify the threshold of nonlinearity [1]. The maximum recorded area under the curve (AUC) and its corresponding single EPSP slope was measured as a ratio to allow between-subject comparisons. D-APV 50 µM was applied at the end of the experiment during ongoing maximal stimulation. The experiment was repeated in the presence of 4AP 0.3 mM, CP339818 5 µM, carbachol 1 µM, or 77-LH-28-1 7 µM.

Experiment: Intrinsic cell properties
Using a potassium-based internal solution, sometimes containing 1 mg/mL neurobiotin, intrinsic cell properties were measured in a specific order. Immediately after breaking in wholecell, the clamping configuration was changed to I=0 to record the resting membrane potential (RMP). Subsequently, the current clamp configuration was adopted for rheobase experiments and enough current was injected to keep the cell at -65 mV. Depolarising 0.8 s current steps of variable magnitude were applied until a single action potential was fired consistently. A chirp current injection followed, 40 pA peak to peak and 20 s in duration, increasing in frequency from 0.2 to 20 Hz, to allow the measurement of impedance and resonance. Subsequently, hyperpolarising and depolarising steps (-150 to 250 pA and 0.8 s). 20 µM ZD-7288 was then washed on, followed by a repeat of the current step and chirp current injection protocols. Impedance was calculated by first applying the fast Fourier transform algorithm to the input current and output voltage sinusoid chirp data and then by taking their complex ratio. Savitzky-Golay filtering was used to smooth the data.

Immunohistochemistry and morphological analysis
Brain slices were placed in 4% PFA for 24-40 hours and then kept in phosphate buffered saline (PBS) (4 °C). Slices were washed in PBS, permeabilised in TritonX100 1:100 in PBS for 1 h, washed and incubated in PBS with 3% bovine serum albumin (BSA) 1 h. Alexafluor-594-streptavidin (1:1000) in PBS with 3% BSA was applied to bind to neurobiotin. Slices were washed in PBS with 3% BSA and incubated with DAPI (1:1000) in PBS with 3% BSA. Slices were then mounted onto glass slides. The slices were imaged using a widefield fluorescence microscope, using Leica LAS X software. The acquired images were processed using Fiji imageJ 1.8.0 software. The publicly available Simple Neurite Tracer plugin was used for semi-automated tracing, visualisation, and analysis of the images [2]. The Sholl Analysis extension of the Simple Neurite Tracer plugin was also employed [3].

Protein Quantification
Western blot analysis was conducted on hippocampal tissue from wt (n=12) and het (n=12) animals. Each tissue sample was lysed in Syn-PER lysis and extraction buffer (Thermo Fisher, UK) with mini protease inhibitor cocktail (Roche Diagnostics) and phosphatase inhibitor (Cell signalling, UK) according to description from manufacturer. After using BCA Assay kit to measure the total amount of protein in each sample, electrophoresis and blotting were carried out. Gels (4-12% NuPAGE Bis-Tris Midi, 45 well) were loaded with 40ug of protein was loaded per well. Samples were added to Laemmli buffer at a 1:1 ratio and this mixture heated at 96 o C for 5 minutes to denature protein-protein interactions and facilitate antibody bindings.
Samples were arranged so that genotypes were counterbalanced across gels with a WT Membranes were then subject to 3 x 10 minute TBST washes before incubation with the appropriate fluorescent IRDye 680RD secondary antibodies at 1:15,000 dilution in 5% milk at room temperature. After another series of TBST washes membranes were imaged on Odyssey CLx Imaging System (Li-COR, Germany). Densiometric analysis of bands was performed using ImageLab 6.0. The densities (with background subtracted) of the protein of interest were divided by the loading control densities for each sample to provide normalised values. Densities were then averaged by group.

Computational modelling
All modelling work was done using the NEURON simulation environment [4]. In-house morphological reconstructions of wt and het neurons were used. A total of 6 representative reconstructions (3 wt, 3 het) were populated with mechanisms corresponding to Ka [5-6], Kir [7][8][9], Km [10], Kdr [5], and NaV [5,10] channels with parameter values tuned to replicate experimental data to assess input resistance in relation to morphology (Table S2). The channel mechanisms were taken from the Neuron Model database [11]. To compare the contributions of the different candidate channels to input resistance, a wt reconstruction was populated with leak current and Ka or Kir channel mechanisms, whose conductance was scaled by 0.5-5. To assess the contribution of input resistance and potassium channels on dendritic integration, the following series of steps were performed. Dendrites were assigned a number which was then shuffled (the same seed was used across conditions). Dendrites were cumulatively recruited using a glutamate mechanism (1 synapse per dendrite), with the ratio of apical proximal and apical tuft dendrites systematically iterated upon. This process was then repeated following changes to conditions, such as the inclusion of a potassium channel mechanism. All dendritic integration simulations were done on the representative wt reconstruction. Model parameters are summarised in Table S3 [12][13][14].   Summary values depicted as mean ± SEM. * P < 0.05, ** P < 0.01, *** P < 0.001 (3-way ANOVA between subject effect)