Abstract
SYNGAP1 haploinsufficiency in humans leads to severe neurodevelopmental disorders characterized by intellectual disability, autism, epilepsy, and sensory processing deficits. However, the circuit mechanisms underlying these disorders are not well understood. In mice, a decrease of SynGAP levels results in cognitive deficits by interfering with the development of excitatory glutamatergic connections. Recent evidence suggests that SynGAP also plays a crucial role in the development and function of GABAergic inhibitory interneurons. Nevertheless, it remains uncertain whether and to what extent the expression of SYNGAP1 in inhibitory interneurons contributes to cortical circuit function and related behaviors. The activity of cortical neurons has not been measured simultaneously with behavior. To address these gaps, we recorded from layer 2/3 neurons in the primary whisker somatosensory cortex (wS1) of mice while they learned to perform a whisker tactile detection task. Our results demonstrate that mice with interneuron-specific SYNGAP1 haploinsufficiency exhibit learning deficits characterized by heightened behavioral responses in the absence of relevant sensory input and premature responses to unrelated sensory stimuli not associated with reward acquisition. These behavioral deficits are accompanied by specific circuit abnormalities within wS1. Interneuron-specific SYNGAP1 haploinsufficiency increases detrimental neuronal correlations directly related to task performance and enhances responses to irrelevant sensory stimuli unrelated to the reward acquisition. In summary, our findings indicate that a reduction of SynGAP in inhibitory interneurons impairs sensory representation in the primary sensory cortex by disrupting neuronal correlations, which likely contributes to the observed cognitive deficits in mice with pan-neuronal SYNGAP1 haploinsufficiency.
SIGNIFICANCE STATEMENT SYNGAP1 haploinsufficiency leads to severe neurodevelopmental disorders. The exact nature of neural circuit dysfunction caused by SYNGAP1 haploinsufficiency remains poorly understood. SynGAP plays a critical role in the function of GABAergic inhibitory interneurons as well as glutamatergic pyramidal neurons in the neocortex. Whether and how decreasing SYNGAP1 level in inhibitory interneurons disrupts a behaviorally relevant circuit remains unclear. We measure neural activity and behavior in mice learning a perceptual task. Mice with interneuron-targeted disruption of SYNGAP1 display increased detrimental neuronal correlations and elevated responses to irrelevant sensory inputs, which are related to impaired task performance. These results show that cortical interneuron dysfunction contributes to sensory deficits in SYNGAP1 haploinsufficiency with important implications for identifying therapeutic targets.
Introduction
De novo loss-of-function variants of the gene SYNGAP1 cause neurodevelopmental disorders characterized by intellectual disability, developmental delay, autism, epilepsy, and sensory processing deficits (Berryer et al., 2013; De Rubeis et al., 2014; Satterstrom et al., 2020). SYNGAP1 encodes a synaptically localized GTPase-activating protein (SynGAP) that enhances the intrinsic GTPase activity of H-Ras and interacts with a synaptic scaffolding protein PSD-95 (Chen et al., 1998; Kim et al., 1998; Komiyama et al., 2002). The importance of SynGAP in neuronal maturation, synapse development, and plasticity is well documented (Komiyama et al., 2002; Kim et al., 2003; Clement et al., 2012; Araki et al., 2015; Barnes et al., 2015; Llamosas et al., 2021). However, it is unclear how pathogenic SYNGAP1 variants impact neural circuits and lead to behavioral abnormalities.
Although SYNGAP1 is predominantly expressed in excitatory neurons of forebrain structures, including the cerebral cortex and hippocampus, its expression is also detected in inhibitory neurons (Zhang et al., 1999; Berryer et al., 2016; Su et al., 2019; Velmeshev et al., 2019). GABAergic inhibitory interneurons are critically important for the regulation of cortical activity and are frequently disrupted in neurodevelopmental disorders (Contractor et al., 2021). The potential therapeutic efficacy of targeting inhibitory neurons has been demonstrated in a mouse model of Fragile-X syndrome (Goel et al., 2018). The question of whether SynGAP functions in inhibitory circuits has received relatively little attention. SynGAP is essential for the migration of inhibitory neurons during development (Su et al., 2019). Pan-neuronal haploinsufficiency of SYNGAP1 impacts inhibitory as well as excitatory neurons (Michaelson et al., 2018; Sullivan et al., 2020). Importantly, a selective loss of SYNGAP1 in GABAergic neurons disrupts the ability of parvalbumin-expressing (PV) inhibitory cortical interneurons to provide perisomatic inhibition in a cell-autonomous manner (Berryer et al., 2016). The resulting loss of inhibition onto pyramidal neurons may contribute to altered cortical γ oscillations and cognitive deficits. However, the circuit-level consequences of reducing SynGAP in GABAergic neurons remain unclear, since neural and behavioral phenotypes have not been analyzed in the same animals.
To investigate how inhibitory interneuron-specific SYNGAP1 haploinsufficiency impacts cortical circuit and sensory perception, we knocked out a copy of the SYNGAP1 gene selectively in vesicular GABA transporter (Vgat)-expressing neurons using Cre-lox. We probed circuit-level alterations using two-photon calcium imaging of the whisker primary somatosensory cortex (wS1), to capitalize on its well-defined circuitry. Mice were trained to detect whisker vibration and indicate it by licking a reward port. This method parallels a previous study connecting SYNGAP1 haploinsufficiency with abnormal processing of whisker input (Michaelson et al., 2018). Importantly, we monitored neural responses in layer 2/3 (L2/3) of wS1 during task performance to assess both behavioral and circuit-level consequences of interneuron-specific SYNGAP1 haploinsufficiency in the same animal.
Interneuron-specific SYNGAP1 heterozygous mice (Vgat-Het) exhibit deficits characterized by elevated behavioral responses in the absence of whisker vibration and premature responses to a sensory input (auditory tone) unrelated to reward acquisition. These behavioral deficits are associated with specific circuit abnormalities within wS1. Pairwise noise correlations in Vgat-Het mice exhibit a broad distribution with higher variability compared with WT mice. Additionally, Vgat-Het mice showed a modest yet statistically significant increase in the magnitude of noise correlations. The elimination of noise correlations had a more pronounced effect on decoding stimulus identity from the population activity of L2/3 neurons in Vgat-Het mice compared with WT mice, leading to improved decoding performance. This suggests that groups of neurons with abnormally elevated correlations may contribute to the impaired behavioral performance of Vgat-Het mice. Furthermore, an increased number of L2/3 neurons in wS1 of Vgat-Het mice responded to the nonrewarded auditory tone, which likely contributes to premature behavioral reports. Collectively, we show that a reduction of SynGAP in inhibitory interneurons results in circuit dysfunction in the primary sensory cortex, characterized by increased detrimental correlations and elevated responses to irrelevant sensory stimuli.
Materials and Methods
Mice
All procedures were in accordance with protocols approved by the University of Michigan Animal Care and Use Committee. We report data from simultaneous calcium imaging and behavior from 5 SYNGAP1 WT mice (VgatCre/+; SYNGAP1+/+) and 5 Vgat-Het mice (VgatCre/+; SYNGAP1fl/+) (The Jackson Laboratory) with a C57BL/J6 background, with ages ranging from 8 to 15 weeks. For experiments characterizing behavioral phenotypes, we used 8 SYNGAP1 WT mice (VgatCre/+; SYNGAP1+/+; 5 of 8 with calcium imaging), 7 Vgat-specific heterozygous mice (VgatCre/+; SYNGAP1fl/+; 5 of 7 with calcium imaging), 5 global SYNGAP1 heterozygous mice (Vgat+/+; SYNGAP1lx-st/+), and 6 GABA-specific heterozygous rescue mice (VgatCre/+; SYNGAP1lx-st/+) (The Jackson Laboratory) with a C57BL/J6 background, with ages ranging from 8 to 15 weeks. Both sexes were used. Mice were housed in a vivarium with a reversed light-dark cycle (12 h each phase). Experiments occurred during the dark phase. After recovery from headpost surgery (see below), mice were singly housed and water-restricted by giving them 1 ml per day. Mouse weight did not decrease below 70% of the starting weight.
Behavioral tasks
Head-restrained mice were trained to perform a Go/No-go whisker detection task using a behavioral apparatus controlled by BPod (Sanworks). Mice were placed in an acrylic (4.5 cm inner diameter) tube. For 7-10 d before training, mice received 1 ml of water per day. Mice were weighed before and after training sessions to measure the amount of water consumed. In the first three sessions (“habituation”), mice were allowed to freely lick at the water port positioned near their snout. Each time the tongue crossed the infrared beam to touch the water port, the mouse received a drop of water (∼7 µl). For training in Go/No-go sessions (7-10 d), facial whiskers were threaded through a plastic mesh attached to a piezoelectric actuator (CTS) and were deflected for 1 s with sinusoidal deflection (rostral to caudal) at 25 Hz on Go trials (60% of trials). On No-go trials (40% of trials), the whiskers were not deflected. One second after trials started, a 0.1 s auditory tone (8 kHz tone, ∼70 dB SPL) was delivered, followed by a 1.5 s No-lick window. If mice lick during this window (i.e., premature licking), the trial was aborted. Licks occurring during the first 0.2 s after the onset of whisker deflection had no consequence. The response window was defined as 0.2-3.2 s after the onset of whisker deflection. Go trials resulted in a “hit” when the mouse licked the water port within the response window and received a drop of water. A “miss” occurred if mice did not lick within the response window, and no reward or punishment was delivered. The No-go trials resulted in a “false alarm” if the mouse licked within the response window, and the mouse was punished by a 3 s timeout. Licking during timeout resulted in an additional timeout. A “correct rejection” occurred if mice did not lick within the response window on No-go trials. During all sessions, ambient white noise (cutoff at 40 kHz, ∼60 dB SPL) was played through a separate speaker to mask any other potential auditory tones associated with the movement of the piezoelectric actuator. No more than three trials of the same type occurred in a row. The fraction of correct trials (Fraction Correct) was defined as the number of hit and correct rejection trials divided by the total number of trials. The hit rate was defined as the number of hits divided by the number of Go trials. The false alarm rate was defined as the number of false alarms divided by the number of No-go trials. The abort rate was defined as the number of aborted trials divided by the number of all trials.
For calcium imaging of inhibitory interneurons in awake, naive mice (see Fig. 7), we applied facial whisker deflection for 1 s using sinusoidal deflection (rostral to caudal) at a randomly selected frequency from 5, 15, 25, or 30 Hz in each trial. Each session consisted of 40-60 trials for each frequency. In subsequent sessions, we recorded tone-evoked responses by delivering a 0.1 s auditory tone (8 kHz tone, ∼70 dB SPL) in each trial. Each session consisted of 40-60 trials of tone presentation.
Surgery and virus injection
Mice were anesthetized with 1% isoflurane throughout surgery and kept on a thermal blanket to maintain body temperature. The scalp and periosteum over the skull were carefully removed. A circular craniotomy was made on the left hemisphere (3.0 mm diameter) with the dura left intact. The center of the craniotomy was located over the wS1 barrel cortex (3.5 mm lateral and 1.3 mm caudal relative to bregma). Injections were performed unilaterally using a beveled glass pipette (30-50 µm diameter) mounted on an oil-based hydraulic micromanipulator (Narishige). Adeno-associated virus (AAV) for expressing jGCaMP6f or jGCaMP7f under the synapsin-1 promoter (AAV1-syn-jGCaMP6f-WPRE-SV40, Addgene, 100837; AAV1-syn-jGCaMP7f-WPRE, Addgene, 104488) was injected into the wS1 at a depth of 250 µm below the dura and at a rate of 1 nl/s (100 nl total). Both WT and Vgat-Het groups contained 3 jGCaMP6f- and 2 jGCaMP7f-injected mice. For GCaMP imaging in Vgat-positive inhibitory interneurons, we injected Cre-dependent GCaMP virus (AAV1-EF1a-DIO-GCaMP6s-P2A-nls-dTomato) to 2 WT mice (VgatCre/+; SYNGAP1+/+) and 2 Vgat-Het mice (VgatCre/+; SYNGAP1fl/+) to record the activity of inhibitory interneurons. The injection was made at three different locations on the cortical surface around the coordinates given above. The craniotomy was covered with a glass window after the injection. The window was made by gluing two pieces of coverslip glass together. The smaller piece (3.0 mm diameter) was placed into the craniotomy while the larger piece (4.0 mm diameter) was glued to the bone surrounding the craniotomy. Cyanoacrylate adhesive (KrazyGlue) and dental acrylic (Jet Repair Acrylic) were used to secure a titanium headpost in place on the skull. Silicone elastomer (Kwik-Cast, WPI) was placed over the window for protection during the recovery period. The mouse was allowed to recover from surgery for at least 10 d before moving to water restriction. Imaging started 3-5 weeks after surgery.
Two-photon calcium imaging of layer 2/3 neurons
Images were acquired on a Scientifica two-photon microscope (Hyperscope) equipped with an 8 kHz resonant scanning module, 2 GaAsP photomultiplier tube modules, and a 16 × 0.8 NA microscope objective (Nikon). jGCaMP was excited at 960 nm (40-60 mW at specimen) with an InSight X3 tunable ultrafast Ti:Sapphire laser (Spectra-Physics). Imaging fields were restricted to areas where jGCaMP expression overlapped with the center of the cranial window (3.5 mm lateral and 1.3 mm caudal to bregma). The beam was focused to 150-250 μm from the cortical surface. The FOV ranged from 458 µm × 344 µm to 275 µm × 207 µm. Images were acquired with a resolution of 512 × 512 pixels at 30 Hz using ScanImage. A movie for a single trial consisted of 140 frames.
Image analysis
Image stacks were processed using the Suite2P pipeline in Python (Stringer and Pachitariu, 2019). After correcting for motion, ROIs were selected and then manually curated to remove ROIs that were not neurons. The neuropil fluorescence time series was multiplied with a correction factor of 0.7 and then subtracted from the raw fluorescence time series to obtain the corrected fluorescence time series: Fcorrected(t) = Fraw − Fneuropil × 0.7. ΔF/Fo was calculated as (F – Fo)/Fo, where Fo represents the baseline fluorescence calculated by determining the average fluorescence (F) in the 8 frames time window preceding whisker stimulus onset. Evoked ΔF/Fo responses were calculated as the average ΔF/Fo over the 10 frames following the expected whisker stimulus onset time. In the case of traces obtained from inhibitory neurons, the baseline fluorescence (Fo) was calculated by averaging fluorescence (F) from the 30 frames preceding the onset of the whisker stimulus. The evoked ΔF/Fo responses were calculated as the average ΔF/Fo over the 30 frames following the onset of the whisker stimulus.
Single-neuron analysis
To quantify the response fidelity, we calculated the percentage of whisker stimulus-responsive trials for each neuron. If the 15 frames (0.5 s) following the onset of whisker stimulus contained three or more frames with fluorescence intensity > (mean over 8 frames preceding the stimulus onset time + 3 × SDs) or < (mean over 8 frames preceding the stimulus onset time – 3 × SDs), the trial was considered responsive. To assign each neuron as “responsive” or “unresponsive,” we used 0.02 as the response fidelity cutoff so that neurons with response fidelity > 0.02 were considered responsive. The receiver operating characteristic (ROC) analysis was used to calculate auROCstim (Kwon et al., 2016). We used across-trial z scores for the evoked ΔF/Fo as a decision variable for each neuron. Trials were grouped by stimulus condition (present vs absent). The ROC curve was computed by systematically varying the criterion value across the full range of decision variables (using the Python sklearn roc_auc_score function). The area under the ROC curve (auROC) represents the performance of an ideal observer in categorizing trials based on the decision variable.
Noise correlation analyses
We calculated across-neuron pairwise noise correlations between neuron pairs recorded at the same time in a single session, across trials sharing the same stimulus condition (the presence or absence of whisker stimulus) (Kwon et al., 2016, 2018). For each neuron, evoked ΔF/Fo responses were first z-scored in Go and No-go trials separately. Pairwise noise correlation was then calculated as the Pearson correlation coefficient between vectors of concatenated z-scored responses for each pair of neurons.
Population decoding analysis
A support vector machine classifier (SVM, Python sklearn package) was trained to discriminate the stimulus condition (Go vs No-go trials) based on a vector of z-scored evoked ΔF/Fo for all responsive cells in each session. Fivefold cross-validation was performed by using a random 80% of trials for training and the remaining 20% for testing the classifier performance. Population decoding accuracy was calculated as the percentage of trials in the test set that was accurately predicted by the classifier. This was done 100 times, and the average accuracy was used for comparison between animals. Go (60%) and No-go (40%) trials were used and randomly assigned to train and test subgroups. To evaluate how noise correlations contribute to neuronal performance, we randomly shuffled trial labels within the same trial type (Go or No-go) in neurons independently from each other by using numpy.random.permutation to remove correlated trial-by-trial variability. Then we used SVM as above to predict stimulus conditions based on the population activity and calculated the average accuracy.
Open field test
Mice were habituated in the open field apparatus (30 × 30 × 30 cm) for 5 min, and then total distance traveled (cm) within 8 min was recorded and analyzed using software (ToxTrac). Each mouse was tested for 2 or three trials, and the distance was averaged across trials.
Statistical tests
The statistical significance of differences between WT and Vgat-Het mice was assessed using unpaired Wilcoxon rank-sum tests unless mentioned otherwise. To test whether WT and Vgat-Het mice form statistically distinct clusters on a scatter plot, we calculated the distance between two centroids, each representing the mean of data points from WT or Vgat-Het mice. We then derived a distribution representing “null hypothesis” by shuffling labels (WT or Vgat-Het) associated with data points 100 times and calculating the distance between centroids each time. If the intercentroid distance is greater than the central 95% of the “null” distribution, WT and Vgat-Het were considered to be separate from each other.
Data availability
Original data reported in this paper will be shared by the lead contact on request. This paper does not report original code. Source data and scripts used for data analysis will be available on request.
Results
Whisker-guided tactile Go/No-go task in head-restrained SYNGAP1 heterozygous mice
Although SYNGAP1 expression has been detected in inhibitory interneurons in the cerebral cortex, there has been no quantitative comparison of its level across major neuronal subtypes. We analyzed SYNGAP1 expression in cortical interneuron subtypes using a published web database (http://research-pub.gene.com/NeuronSubtypeTranscriptomes) that includes the transcriptome dataset collected by the Allen Brain Institute (Tasic et al., 2016; Huntley et al., 2020). SYNGAP1 expression was detected in all three major inhibitory interneuron subtypes and excitatory cells in the cortex, consistent with previous studies (Fig. 1A) (Zhang et al., 1999). To generate inhibitory interneuron-specific SYNGAP1 heterozygous (VgatCre/+;SYNGAP1fl/+ or simply “Vgat-Het”) and control (VgatCre/+;SYNGAP1+/+ or “WT”) mice, we crossed “floxed” SYNGAP1 mice (SYNGAP1fl/+) with Vgat-IRES-Cre (Vgatcre/cre) mice (Fig. 1B). The mouse line carrying SYNGAP1fl/fl was previously reported (Ozkan et al., 2014).
Tactile detection task in interneuron-specific SYNGAP1 heterozygous mice. A, Expression of SYNGAP1 transcripts in excitatory (Exc), and inhibitory — PV, SST, and vasointestinal peptide (VIP) — neurons based on an analysis of a single-cell transcriptomics dataset. Transcript level was quantified as reads per kilobase of transcript, per million mapped reads (RPKM). B, Breeding scheme for knocking out a copy of SYNGAP1 in inhibitory interneurons. SYNGAP1fl/+ mice were crossed with Vgat-IRES-Cre to create VgatCre/+; SYNGAP1fl/+ mice (Vgat-Het). VgatCre/+; SYNGAP1+/+ mice were used as the WT control. C, Schematic represents the experimental setup. Head-fixed mice were trained to lick a water port if facial whiskers were sinusoidally vibrated at 25 Hz with a piezoelectric stimulator or to withhold licking in the absence of whisker vibration. Neural responses in L2/3 of wS1 were monitored using two-photon imaging of genetically encoded calcium indicator jGCaMP6f or jGCaMP7f. D, Experimental timeline. P, Postnatal. E, Each trial began with a brief auditory tone and was subsequently followed by 1.5 s No-lick window; licking during this period resulted in trial abortion. In 60% of trials (Go), the whiskers were then vibrated at 25 Hz for 1 s. In 40% of trials (No-go), the whiskers were not vibrated. F, Trial structure and outcomes. G, Example post-training sessions from WT and Vgat-Het animals, showing individual licks in a veridical trial order (left) and sorted by trial type (right). Ticks indicate lick times assorted by five different trial outcomes: hit (black), miss (cyan), false alarm (red), correct rejection (blue), and aborted (magenta) trials.
Inspired by prior work demonstrating altered tactile processing in the pan-neuronal SYNGAP1 heterozygous KO (Michaelson et al., 2018; Llamosas et al., 2021), we first tested whether or not interneuron-specific SYNGAP1 haploinsufficiency impacts tactile perception using a head-fixed whisker detection task (Fig. 1C). Adult mice were water-restricted, habituated, and then trained to report, by licking or withholding licking of a reward port, whether facial whiskers received a brief sinusoidal deflection (25 Hz for 1 s, peak speed ∼800 degrees s−1) (Fig. 1D). Each trial began with a brief auditory tone (0.1 s, 8 kHz tone, 70 dB). This was immediately followed by 1.5 s No-lick window, and licking during this period aborted the trial (Fig. 1E). At 2 s after the offset of the auditory tone, the whisker deflection was delivered in 60% of all trials (Go trials). The response window was defined as 0.2-3.2 s after the onset of whisker deflection. Go trials resulted in a hit when the mouse licked the water port within the response window and received a drop of water. In the remainder of trials (No-go trials; 40%), whiskers were not deflected. Licking during the response window in the absence of whisker stimulus resulted in a 3 s timeout. The probability of correct choices during Go (presence of whisker stimulus) and No-go (absence of whisker stimulus) trials was monitored across training. Trial outcomes comprised a mixture of successful detection (hits) and failed detection (misses) following Go trials, as well as correct behavioral responses (correct rejections) and incorrect responses (false alarms) following No-go trials (Fig. 1F). For each trial, individual licks made by the animal were recorded (Fig. 1G).
Heterozygous KO of SYNGAP1 in inhibitory interneurons impairs task learning
WT littermate control mice (VgatCre/+; SYNGAP1+/+) steadily improved their behavioral performance and became “expert” (discriminability index d′ > 2) at the whisker Go/No-go detection task after 6 or 7 daily sessions, whereas the performance of Vgat-Het mice (VgatCre/+; SYNGAP1fl/+) hovered between 1 < d′ < 2 (Fig. 2A). Performances of WT and Vgat-Het mice were comparable on Session 1 (pre-training), but WT mice performed significantly better than Vgat-Het mice on Session 7 (post-training; p = 0.014) (Fig. 2B). The difference in post-training performance (Session 7) was driven by higher false alarm rates of Vgat-Het mice, although they also displayed a slightly lower hit rate (hit rate: p = 0.174; false alarm rate: p = 0.013) (Fig. 2C, D, left). The response criterion was not altered in Vgat-Het mice, indicating that their tendency to respond in the presence or absence of stimulus is comparable to that of WT mice (Fig. 2D, right). Importantly, the fraction of aborted trials was elevated in Vgat-Het compared with WT mice in Sessions 6 and 7, indicating an increased number of premature licks made during the No-lick window (p = 0.0205) that precedes whisker stimulus onset (Fig. 2E). The amount of water consumption (Fig. 2F) or locomotor activity measured using open-field exploration (Fig. 2G) was not altered in Vgat-Het mice (p = 0.090; p = 0.097). Therefore, the increased number of aborted trials cannot be accounted for by differences in motivation or general locomotor activity.
Vgat-Het mice show aberrant behavioral responses to sensory input. A, Performance (d′) as a function of training session for WT (black) and Vgat-Het (red) animals. Error bars indicate ± SEM from 8 WT and 7 Vgat-Het mice. *p = 0.014. **p = 0.004. Two-way repeated-measures ANOVA with Bonferroni’s multiple comparison, Genotype: F(1,91) = 17.17, p < 0.0001; Session: F(6,91) = 1.413, p = 0.2184; Genotype × Session interaction: F(6,91) = 1.691, p = 0.1320. B, Performance (d′) of individual WT (black) and Vgat-Het (red) mice on Session 1 (pretraining) and Session 7 (post-training). Error bars indicate ± SEM from 8 WT and 7 Vgat-Het mice. *p = 0.014. C, Hit rate for WT (black) and Vgat-Het (red) animals plotted over the course of training sessions. Error bars indicate ± SEM from 8 WT and 7 Vgat-Het mice. Two-way repeated-measures ANOVA with Bonferroni’s multiple comparison, Genotype: F(1,91) = 1.980, p = 0.1627; Session: F(6,91) = 0.7654, p = 0.5990; Genotype × Session interaction: F(6,91) = 0.7739, p = 0.5925. D, Left, False alarm rate for WT (black) and Vgat-Het (red) animals plotted over the course of training sessions. Error bars indicate ± SEM from 8 WT and 7 Vgat-Het mice. *p = 0.013. Two-way repeated-measures ANOVA with Bonferroni’s multiple comparison, Genotype: F(1,91) = 4.702, p = 0.0327; Session: F(6,91) = 2.995, p = 0.0102 ; Genotype × Session interaction: F(6,91) = 0.4337, p = 0.8547. Right, Response criterion for WT (black) and Vgat-Het (red) animals plotted over the course of training sessions. Error bars indicate ± SEM from 8 WT and 7 Vgat-Het mice. p = 0.2319. Two-way repeated-measures ANOVA with Bonferroni’s multiple comparison, Genotype: F(1,91) = 2.802, p = 0.0976; Session: F(6,91) = 0.6904, p = 0.6579 ; Genotype × Session interaction: F(6,91) = 0.6197, p = 0.7140. E, Average fraction of aborted trials as a function of training sessions for WT (black) and Vgat-Het (red) animals. Error bars indicate ± SEM from 8 WT and 7 Vgat-Het mice. *p = 0.0205. Two-way repeated-measures ANOVA with Bonferroni’s multiple comparison, Genotype: F(1,91) = 4.346, p = 0.0399; Session: F(6,91) = 5.757, p < 0.0001; Genotype × Session interaction: F(6,91) = 0.7800, p = 0.5877. F, Water consumption of each day. Error bars indicate ± SEM from 8 WT and 8 Vgat-Het mice. p = 0.090. G, Open field test distance traveled within 8 min. Error bars indicate ± SEM from 7 WT and 7 Vgat-Het mice. p = 0.097. H, Breeding scheme for rescuing a copy of SYNGAP1 in inhibitory interneurons. Cre-conditional rescue mouse carrying lox-stop (SYNGAP1lx-st/+) was crossed with Vgat-IRES-Cre to create VgatCre/+; SYNGAP1lx-st/+ mice (Vgat-Res). I, Hit rate of individual WT (black; n = 8), Vgat-Het (red; n = 7), Het (blue; n = 5) mice, and Vgat-Res (purple; n = 6) in Session 1 (pretraining) and Session 7 (post-training). Error bars indicate ± SEM. J, False alarm rate in Session 1 (pretraining) and Session 7 (post-training). Error bars indicate ± SEM. *p = 0.013 (WT against Vgat-Het). **p = 0.002 (WT against Het). *p = 0.0303 (Het against Vgat-Res). K, Fraction of correct trials in prelearning and post-learning sessions, calculated as (Hits + Correct Rejection)/all trials, including aborted ones. ***p = 0.0004 (WT against Vgat-Het). *p = 0.009 (WT against Het). *p = 0.0303 (Het against Vgat-Res). L, Abort rate on prelearning and post-learning sessions, calculated as (Hits + Correct Rejection)/all trials, including aborted ones. *p = 0.0205 (WT against Vgat-Het). *p = 0.0210 (WT against Het). *p = 0.0173 (Het against Vgat-Res).
To what extent does interneuron-targeted SYNGAP1 disruption contribute to behavioral deficits observed in global SYNGAP1 heterozygous mice? To address this, we tested (1) global SYNGAP1 heterozygous (Het) mice (SYNGAP1lx-st/+) (Clement et al., 2012) and (2) inhibitory interneuron-specific SYNGAP1 rescue (Vgat-Res) generated by crossing Cre-conditional rescue carrying lox-stop (SYNGAP1lx-st/+) with Vgat-IRES-Cre (Vgatcre/cre) (Fig. 2H). Post-learning hit rates of Het and Vgat-Res mice were comparable to the WT (Fig. 2I). Consistent with prior studies, Het mice displayed significantly elevated post-training false alarm rates (p = 0.002) compared with WT (Fig. 2J) (Michaelson et al., 2018). Vgat-Het and Het showed similar post-training false alarm rates. Interestingly, normalizing SYNGAP1 expression in inhibitory interneurons (Vgat-Res) rescued the elevated false alarm rate (p = 0.0303) (Fig. 2J). The fraction of correct trials (hit and correct rejection) was significantly reduced by similar magnitudes in Vgat-Het (p = 0.0004) and global Het mice (p = 0.009) and was rescued in Vgat-Res (p = 0.0303) (Fig. 2K). Post-learning abort rates were also significantly higher in both Vgat-Het and global Het mice compared with WT (p = 0.0210); higher abort rate was corrected in Vgat-Res (p = 0.0173) (Fig. 2L). Together, these results demonstrate: (1) impaired sensory learning in Vgat-Het mice, which was driven by elevated behavioral responses in the absence of relevant sensory stimulus, (2) a decrease of SynGAP in GABAergic neurons is sufficient to cause aberrant behavioral responses and sensory learning impairments observed in global SYNGAP1 haploinsufficiency, and (3) normalizing SynGAP level in GABAergic neurons rescues some of these behavioral impairments.
To test how behavioral impairments manifest at the level of motor output, we monitored individual licks made by the animal during task performance (Fig. 1G) and compared their distribution across trials in WT versus Vgat-Het mice before and after training for an equivalent number of sessions (Fig. 3A). Compared with pretraining sessions, WT animals exhibited a decreased fraction of tone-evoked licks in post-training sessions, whereas the number of licks made at the expected time of whisker stimulus onset increased (Fig. 3A). Therefore, WT mice learned to effectively respond to reward-associated sensory input by shifting the time window of licking. Vgat-Het mice, on the other hand, continued to respond to the tone even in Session 7 (post-training) (Fig. 3A). We also compared distributions of time points when the first lick was made after the auditory tone in each trial. First lick times, combined across all trials, were clustered around the expected time of whisker stimulus onset in expert WT mice, whereas they were more distributed in Vgat-Het mice (p < 0.0005) (Fig. 3B). Next, we compared lick distribution in trials excluding aborted trials (Fig. 3C). WT mice again showed more prominent training-induced increases in licking immediately after the expected time of whisker stimulus onset, compared with Vgat-Het mice (Fig. 3C). We calculated the ratio between licks made within a 1 s time window after whisker stimulus onset versus licks outside this window (Fig. 3D). Through training, both WT and Vgat-Het mice increased the number of licks within 1 s of whisker stimulus onset (p < 0.0005), but the change was significantly greater in WT mice (p < 0.0005) (Fig. 3D). The distribution of pretone licks in No-go trials was similar between WT and Vgat-Het mice (Fig. 3E), suggesting that the increased number of false alarm trials in Vgat-Het mice is attributed to an elevated number of licks in the absence of whisker stimulus, rather than a general increase in lickiness. Regarding the aborted trials, both WT and Vgat-Het exhibited an increase in the number of licks following the tone onset (Fig. 3F). This finding indicates that the elevated number of aborted trials is a result of an increase in tone-evoked licks, rather than elevated occurrence of nonspecific licks throughout the trials.
Vgat-Het mice exhibit altered patterns of motor responses to the sensory input. A, Average proportion of licks within 200 ms time bins, calculated as a fraction over the total number of licks combined across all trials, including aborted ones. Red dashed line indicates the onset of auditory tone. Black dashed line indicates the expected onset of whisker stimulus. Error bars indicate ± SEM from 8 WT and 7 Vgat-Het mice. B, Distribution of the first lick times following the auditory tone during pretraining and post-training sessions of WT versus Vgat-Het animals. Box plot represents median and IQR. Well-trained WT mice (orange) make their first licks tightly around the expected onset time of whisker vibration, whereas a significant fraction of first licks made by post-training Vgat-Het mice (magenta) occurs before the expected onset of whisker stimulus. ***p < 0.0005. C, Same as in A, but excluding aborted trials. D, Fraction of licks that occur within 1 s (C, gray shade) after the expected onset of whisker stimulus. Box plot represents median and IQR. WT animals (orange) target their licks more efficiently to the learned reward cue (within 1 s after the onset of whisker stimulus), compared with Vgat-Het mice (magenta). ***p < 0.0005. E, Same as in A, but only showing No-go trials. F, Same as in A, but only showing aborted trials.
In summary, Vgat-Het mice show heightened behavioral responses to the irrelevant sensory input (auditory tone) and during the time window when the relevant stimulus (whisker vibration) is absent, compared with WT mice. Our results suggest that SYNGAP1 haploinsufficiency in GABAergic neurons impairs learning by driving impulsive behavioral responses during a goal-directed tactile perception task.
Neuronal sensitivity to whisker stimulus is slightly reduced in Vgat-Het mice
Next, we characterized the nature of the cortical dysfunction that underlies the behavioral impairments described above. To evaluate the relationship between behavioral performance and neural representations, we monitored responses of layer 2/3 neurons in the wS1 with two-photon calcium imaging of a genetically encoded calcium indicator (jGCaMP6f or 7f) as mice learned to perform the detection task (Fig. 4A,B). We reasoned that, if SYNGAP1 plays a role in the inhibitory circuit in the wS1, reducing its expression in cortical inhibitory interneurons should alter whisker input representation.
Whisker input representation by L2/3 neurons is impaired in Vgat-Het mice. A, Example FOV in the two-photon microscope, showing jGCaMP7f-expressing neurons. Scale bar, 100 µm. B, Representative calcium transients evoked by whisker stimulation in Go trials. Different colors correspond to different cells as indicated by colored circles in A. Dashed lines indicate the onset of whisker deflection. C, Top, Magnitude of whisker-evoked responses in all cells on Go trials before and after training. Median [IQR for WT-pre: 0.022 [–0.089, 0.116]; WT-post: 0.033 [–0.056, 0.123]; Vgat-Het-pre: –0.004 [–0.142, 0.115]; Vgat-Het-post: –0.017 [–0.211, 0.153]. **p = 0.002. ***p < 0.0005. Bottom, Cumulative histograms of mean whisker-evoked ΔF/Fo. D, Bottom, Each line corresponds to the average response of a single whisker-responsive neuron in Go trials in the post-training sessions. ΔF/Fo traces were z-scored across all trials for each neuron and then averaged within Go trials. Neurons were pooled across animals and sorted by the time of peak activity. x axis indicates time in relation to the onset of the whisker stimulus. Top, z-scored average ΔF/Fo traces from post-training sessions in WT and Vgat-Het mice. Error bars indicate ± SEM from 5 WT (orange) and 5 Vgat-Het (magenta) mice. E, Magnitude of whisker-evoked responses in whisker-responsive cells in Go trials. Median [IQR] for WT-pre: 0.045 [–0.051, 0.147]; WT-post: 0.060 [–0.026, 0.154]; Vgat-Het-pre: 0.039 [–0.096, 0.155]; Vgat-Het-post: 0.061 [–0.109, 0.221]. *p = 0.038. F, Left, Representative traces of a neuron with high auROCstim. Middle, Area under receiver-operating-characteristic curve (auROCstim) quantifying how well an ideal observer could categorize the presence or absence of whisker stimulus (Go vs No-go trials) on the basis of the neural response. Median [IQR] for WT: auROCstim-pre = 0.517 [0.468, 0.574], auROCstim-post = 0.528 [0.471, 0.585]; Median [IQR] for Vgat-Het: auROCstim-pre = 0.515 [0.434, 0.579], auROCstim-post = 0.500 [0.375, 0.602]. *p = 0.0489. ***p < 0.0005. Right, Cumulative histograms of auROCstim.
We expressed jGCaMPs under pan-neuronal synapsin 1 promoter in the wS1 by virus injection and implanted a cranial window to enable optical access. We recorded activity from 613 neurons (5 WT mice) and 738 neurons (5 Vgat-Het mice) in the L2/3 of wS1. The average magnitude of neuronal response (ΔF/Fo) evoked by whisker stimulation modestly decreased in Vgat-Het compared with WT mice (pretraining: p = 0.002; post-training: p < 0.0005) (Fig. 4C). We found 41.0 ± 6.54% of WT neurons and 35.3 ± 3.67% of Vgat-Het neurons imaged in wS1 to be responsive to whisker stimulation, indicating a slight reduction in the pool of whisker-responsive L2/3 neurons in Vgat-Het mice. When the analysis was restricted to whisker-responsive cells, however, WT and Vgat-Het mice showed comparable magnitude and temporal dynamics of calcium transients (Fig. 4D). Magnitudes of evoked ΔF/F0 among whisker-responsive cells were similar between WT and Vgat-Het mice (pre-training: p = 0.038; post-training: p = 0.114) (Fig. 4E).
To quantify neuronal sensitivity to the whisker stimulus, we calculated the area-under the receiver-operating-characteristic curve (auROCstim), which captures how well an ideal observer could categorize sensory stimulus (in our case, the presence or absence of whisker deflection) based on the neural response (Kwon et al., 2016). We used stimulus-evoked ΔF/Fo as a decision variable for individual trials. auROCstim significantly increased through training in WT mice (pretraining vs post-training: p = 0.0489), whereas it remained unchanged in Vgat-Het mice (p = 0.0661) (Fig. 4F). auROCstim was also significantly greater in WT mice compared with Vgat-Het mice that went through an equivalent number of training sessions (p < 0.0005) (Fig. 4F). Our results suggest that SYNGAP1 haploinsufficiency in cortical interneurons causes subtle yet significant decreases in neuronal sensitivity to whisker stimulus in the L2/3 of wS1 during training.
Vgat-specific SYNGAP1 KO disrupts population coding of whisker stimulus
Correlated trial-to-trial fluctuation in stimulus-evoked responses severely limits the amount of sensory information encoded by a neuronal population in the cortical network. The correlated covariability between pairs of neurons or noise correlation is usually a small positive number (range: 0.05-0.25) (Cohen and Kohn, 2011) and reflects synaptic connectivity or shared input (Ko et al., 2011). Altered pairwise noise correlations have been observed in several mouse models of neurodevelopmental disorders (Banerjee et al., 2016; Antoine et al., 2019; Lazaro et al., 2019). We calculated pairwise noise correlations among whisker-responsive neurons using their stimulus-evoked ΔF/Fo in individual trials (Kwon et al., 2016, 2018). Noise correlations decreased through learning in both WT and Vgat-Het mice (p < 0.0005 for both comparisons), consistent with the learning-associated improvement in sensory encoding (Fig. 5A). In trained mice, noise correlation was elevated (p = 0.0016) and more widely distributed in Vgat-Het compared with WT (Fig. 5A). To compare the width of the distribution between mice, we calculated the interquartile range (IQR) of noise correlations in individual mice and found that the IQR was significantly larger in Vgat-Het mice (p = 0.015) (Fig. 5B). We interpret this result as a greater number of positively and negatively correlated neuronal pairs being present in Vgat-Het mice compared with WT (Harris and Thiele, 2011).
Neuronal correlations are altered in Vgat-Het mice. A, Pairwise noise correlation calculated using all trials except the aborted ones. Median [IQR] for WT-pre: 0.067 [–0.013, 0.175]; WT-post: 0.023 [–0.027, 0.083]; Vgat-Het-pre: 0.063 [–0.064, 0.218]; Vgat-Het-post: 0.032 [–0.048, 0.120]. *p = 0.019. **p = 0.0016. ***p < 0.0005. B, IQR of distribution of noise correlations. Each dot represents a median IQR of a mouse. *p = 0.015. C, Top, Relationship between animal’s task performance and the distribution of pairwise noise correlation quantified using IQR. Dots indicate individual mice. Error bars indicate ± 95% CI. p = 0.0407. Distance between WT and Vgat-Het centeroids: 0.144; 95% CI of intergroup distance of null distribution is [0.049, 0.053]. Bottom, Relationship between animal’s task performance and the distribution of pairwise noise correlation quantified using SD. Dots indicate individual mice. Error bars indicate ± 95% CI. p = 0.0584. Distance between WT and Vgat-Het centeroids: 0.143; 95% CI of intergroup distance of null distribution is [0.078, 0.147]. D, Testing the impact of noise correlation on accuracy of decoding stimulus conditions from population activity. The SVM classifier was trained to predict stimulus conditions on each trial based on the population activity. Bar graphs represent average decoding accuracy before and after removing noise correlation by shuffling trial labels within both Go and No-go trials (solid) or within Go trials only (hashed). Each dot represents a mouse. Shuffle within Go and No-go: *p = 0.0121; Shuffle within Go: *p = 0.0122, paired t tests.
How does the altered distribution of noise correlations impact an animal’s behavior and/or encoding of sensory information in Vgat-Het mice? To begin to answer this question, we plotted the IQR or SD of noise correlations against performance for individual mice (i.e., the fraction of correct trials) (Fig. 5C). There was a negative relationship between the distribution of noise correlations and task performance (IQR: R2 = 0.4267, p = 0.0407; SD: R2 = 0.3783, p = 0.0584); WT and Vgat-Het mice formed distinct clusters on this plot (Fig. 5C). This suggests that the presence of aberrant noise correlations is likely to contribute to the impaired task performance in Vgat-Het mice. Next, we compared the amount of sensory information encoded by the L2/3 neuronal population. We decoded the stimulus condition (presence or absence of whisker deflection) using an SVM-based classifier trained on evoked ΔF/Fo of the L2/3 neurons. The stimulus condition in individual trials could be predicted with ∼80% accuracy in both WT and Vgat-Het mice (Fig. 5D). We then removed noise correlations by shuffling trial labels in the same trial type (Go and No-go) and asked whether the decoding of stimulus information could be improved. Removing noise correlations in both Go and No-go trials had mixed effects in WT mice with a slight improvement in decoding accuracy that did not meet statistical significance (Go and No-go: p = 0.3417; Go only: p = 0.475) (Fig. 5D). For Vgat-Het mice, on the other hand, removing noise correlations in both Go and No-go trials or Go trials significantly improved the decoding accuracy (Go and No-go: p = 0.0121; Go only: p = 0.0122) (Fig. 5D). Therefore, a decrease of SYNGAP1 expression in GABAergic interneurons introduces aberrant information-limiting noise correlations in the L2/3 population, which reduces the amount of encoded sensory information.
Vgat-specific SYNGAP1 KO increases tone-evoked responses in wS1
Compared with WT mice, Vgat-Het mice lick more in response to the auditory tone at the beginning of trials, resulting in an increased number of aborted trials (Fig. 2E). We hypothesized that neuronal responses to the auditory tone might be abnormally elevated in Vgat-Het mice. To test this, we compared the activity of L2/3 wS1 neurons around the onset of the auditory tone. We focused on trials that do not contain licks immediately following the onset of tone presentation to exclude confounding effects of increased licking (Fig. 6A,B).
L2/3 neurons in wS1 of Vgat-Het mice show enhanced responses to the auditory tone. A, Each line corresponds to a single neuron. ΔF/Fo traces were z-scored first across all trials for each neuron and then averaged within nonaborted trials. x axis indicates time in relation to the onset of whisker stimulation. White dashed line indicates the onset of auditory tone. B, Representative fluorescence traces of auditory tone-responsive neurons. Orange represents WT; magenta represents Vgat-Het. C, Cumulative density plots showing fidelity of tone-evoked responses in the L2/3 neurons of wS1 before and after training. ***p < 0.0005, Kolmogorov–Smirnov test. D, Relationship between abort rate and fidelity of tone-evoked responses. R2 = 0.2284; p = 0.1624. Distance between WT and Vgat-Het centroids: 0.109; 95% CI of intergroup distance of null distribution: [0.038, 0.042]. E, Left, Mean tone-evoked responses in all cells of nonaborted trials. Median [IQR] for WT-pre: –0.013 [–0.073, 0.037];WT-post: –0.006 [–0.035, 0.019]. *p = 0.034; Vgat-Het-pre: 0.00 [–0.050, 0.046]; Vgat-Het-post: 0.005 [–0.036, 0.047], *p = 0.029. **p = 0.0014. ***p < 0.0005. Right, Cumulative histograms of mean tone-evoked ΔF/Fo.
The representation of whisker input in inhibitory cells is impaired in Vgat-Het. A, Left, Example FOV in the two-photon microscope, showing jGCaMP6s expression in inhibitory neurons. Scale bar, 100 μm. Right, Representative calcium transients elicited by whisker stimulation in awake, naive mice. Each color represents a different cell, as indicated by corresponding colored circles in the left panel. Dashed lines indicate the onset of whisker deflection. B, Each line indicates the average calcium response from a single neuron (WT: 109 cells; Vgat-Het: 73 cells). The x axis indicates time within each trial. White dashed line indicates the onset of whisker stimulation. The example shown is from trials with 25 Hz whisker stimulation. C, Magnitude of whisker-evoked responses in inhibitory cells with 5 Hz (p = 0.0031, Kolmogorov–Smirnov test), 15 Hz (p < 0.0005, Kolmogorov–Smirnov test), 25 Hz (p < 0.0005, Kolmogorov–Smirnov test), and 30 Hz (p = 0.0337, Kolmogorov–Smirnov test). D, Each line indicates the average calcium response from a single neuron (WT: 97 cells; Vgat-Het: 70 cells). The x axis indicates time within each trial. White dashed line indicates the onset of auditory tone (0.1 s, 8 kHz tone, 70 dB). E, Magnitude of tone-evoked responses in inhibitory cells. p = 0.0532, Kolmogorov–Smirnov test.
The fidelity of auditory tone-evoked responses was quantified as the fraction of trials in which statistically significant responses were elicited following the tone presentation. The auditory response fidelity was significantly elevated in Vgat-Het compared with WT mice in both pretraining and post-training sessions (p < 0.0005, Kolmogorov–Smirnov test) (Fig. 6C). We then asked whether auditory response fidelity is elevated in mice showing higher abort rates, by plotting the trial abort rate against average auditory response fidelity for individual mice. We found a positive correlation between these two metrics (R2 = 0.2284; p = 0.1624), although it was not statistically significant (Fig. 6D). Analysis of the distance between Vgat-Het and WT data points showed that they formed distinct clusters; Vgat-Het mice show elevated auditory response fidelity and an increased number of aborted trials compared with WT mice. We also compared the average magnitude of tone-evoked ΔF/Fo between Vgat-Het and WT mice. The response magnitude was significantly greater in Vgat-Het compared with WT mice in both pretraining and post-training sessions (pre: p = 0.0014, post: p < 0.0005) (Fig. 6E). We conclude that a larger proportion of L2/3 neurons in the wS1 of Vgat-Het mice respond to the auditory tone. However, the exact impact of the increased tone-evoked response in wS1 on the animal’s performance in the whisker detection task is yet to be determined. It is possible that the heightened responsiveness to tones in wS1 during pretraining sessions could interfere with task-learning.
Vgat-specific SYNGAP1 KO decreases whisker-evoked responses in inhibitory neurons
Up to this point, the experiment has involved all L2/3 neurons in the wS1 of WT and Vgat-Het mice. However, it is important to determine the contribution of inhibitory interneurons in Vgat-Het mice of the deficits in the whisker input representation and neuronal correlations. To address this, we conducted two-photon calcium imaging specifically targeting inhibitory neurons within the wS1. We injected AAVs expressing Cre-dependent GCaMP (AAV1-EF1a-DIO-GCaMP6s-P2A-nls-dTomato) into wS1 of 2 WT (VgatCre/+; SYNGAP1+/+) and 2 Vgat-Het mice (VgatCre/+; SYNGAP1fl/+) that have Vgat-IRES-Cre background (Fig. 7A). We applied sinusoidal deflection of facial whiskers for 1 s at a randomly selected frequency from 5, 15, 25, or 30 Hz in each trial while imaging calcium transients in awake, naive mice. The 25 Hz stimulus was equivalent to what was used in the Go/No-go detection task.
We observed a significant difference in the amplitude of whisker-evoked responses between inhibitory neurons of WT mice and those in Vgat-Het mice. While the whisker deflection at 25 Hz elicited a robust response in the majority of WT neurons, it evoked only a minimal response in Vgat-Het neurons (Fig. 7B). The same trend was observed across the range of whisker stimulus frequency, with WT neurons exhibiting a significantly larger magnitude at 5, 15, 25, and 30 Hz (Fig. 7C). This finding indicates that SYNGAP1 haploinsufficiency in inhibitory neurons of the wS1 makes them less responsive to whisker input, which is likely to cause unchecked aberrant correlations among the neuronal population (Fig. 5). To investigate potential alterations in noise correlations among inhibitory interneurons in Vgat-Het mice, we computed the noise correlations between pairs of whisker-responsive inhibitory neurons (WT: 66 of 110 cells; Vgat-Het: 36 of 74 cells). Our analysis revealed no statistically significant difference in the magnitude of noise correlations between WT and Vgat-Het groups (median [IQR] for WT: 0.303 [0.065, 0.520]; Vgat-Het: 0.299 [0.020, 0.514]; Wilcoxon rank-sum test, p = 0.126). Subsequently, we asked whether the elevated tone-evoked responses in Vgat-Het mice could be attributed to hypoactivity of inhibitory neurons within the wS1. We measured the tone-evoked responses in the inhibitory neurons within wS1 of both WT and Vgat-Het mice during the presentation of a 0.1 s auditory tone (8 kHz, ∼70 dB SPL) in each trial. The same tone was used in the Go/No-go detection task. No significant difference was observed in the amplitude of tone-evoked responses between inhibitory neurons of WT mice and those in Vgat-Het mice (Fig. 7D,E). This finding suggests that a disinhibitory deficit is unlikely to be a major contributor to the increased tone-evoked neuronal response in Vgat-Het mice.
Discussion
Previous studies demonstrated the role of SynGAP in regulating the development and plasticity of excitatory neurons in the neocortex (Clement et al., 2012; Ozkan et al., 2014). More recently, it has been reported that SynGAP also controls the migration and connectivity of cortical inhibitory interneurons (Berryer et al., 2016; Su et al., 2019; Sullivan et al., 2020). In the present study, we tested if and to what extent SYNGAP1 haploinsufficiency in GABAergic cells contributes to behavioral and cortical circuit abnormalities.
Experiments in head-fixed mice performing whisker-guided sensory detection tasks are widely used for probing circuit dysfunction in autism models, including pan-neuronal SYNGAP1 and interneuron-specific SHANK3 KO mice (Michaelson et al., 2018; Chen et al., 2020). Head-fixed preparations were also adopted in studies examining visual processing in autism models (Batista-Brito et al., 2017; Goel et al., 2018; Del Rosario et al., 2021). In most previous studies, neural activity and task performance were not measured simultaneously, however (but see Del Rosario et al., 2021). By combining two-photon calcium imaging and quantitative behavioral tasks in the same animal, we directly tested the relationships between neural activity and behavior in GABAergic SYNGAP1 haploinsufficiency.
Knocking out a copy of SYNGAP1 in Vgat-positive inhibitory interneurons was sufficient to cause learning deficits characterized by impaired detection task performance and increased tendency to respond in the absence of relevant sensory input (Fig. 2). We also confirmed that global SYNGAP1 heterozygous KO impaired task performance and elevated the false alarm rate as previously reported (Michaelson et al., 2018). An increased false alarm rate during a goal-directed sensory perception task was also observed in models of other neurodevelopmental disorders, including Fmr1 KO mice (Goel et al., 2018). A novel finding of this study is that both global and Vgat-specific SYNGAP1 KO mice show an increased number of premature responses to sensory input unrelated to reward acquisition, which resulted in higher rate of trial abortion (Fig. 2). The elevated trial abortion was rescued by normalizing SYNGAP1 in Vgat-expressing neurons. Based on these results, we conclude that SYNGAP1 expression in inhibitory interneurons is required for generating appropriate behavioral responses to sensory input during goal-directed behaviors. It is worth noting that certain mouse models of neurodevelopmental disorders, such as CNTNAP2 KO, have reported decreases in false alarm rates, which are associated with diminished cortical excitation and elevated inhibition (Lazaro et al., 2019; Del Rosario et al., 2021). These findings from the present study and previous research underscore the importance of considering the heterogeneity of circuit mechanisms and neuronal cell types when developing effective therapeutics for neurodevelopmental disorders.
The behavioral deficits described above were accompanied by a disrupted representation of whisker input in the L2/3 of wS1, characterized by the following: (1) reduced neuronal sensitivity to whisker stimulus (auROCstim) (Fig. 4); (2) an increased amount of information-limiting correlations (Fig. 5); and (3) elevated responses to sensory input unrelated to reward acquisition, such as the auditory tone (Fig. 6). Noise correlations have been previously measured in several mouse models of NDDs. Mean noise correlations were found to be decreased in Fmr1 KO (Antoine et al., 2019), decreased in PV neuron-specific MeCP2 KO, but increased in somatostatin (SST)-specific MeCP2 KO (Banerjee et al., 2016). They were unchanged in wS1 of CNTNAP2 KO (Antoine et al., 2019) and altered in the PFC differently depending on neuronal subtypes (Lazaro et al., 2019). However, until now, it has been unclear to what extent altered noise correlations contribute to altered task performance in these mouse models. We have observed that noise correlations among neurons in the primary somatosensory cortex of Vgat-Het exhibit a wider distribution, and this increased variability of noise correlations is associated with reduced task performance.
Whether the wider distribution of noise correlations can be attributed to an increased variability in the strength of local synaptic connections is yet to be determined. The variance of noise correlations has been generally overlooked in previous studies. In one study, they have reported that the aroused state, as indicated by increased pupil diameter, reduces the variance of noise correlations in L2/3 of primary auditory cortex (Lin et al., 2019). This suggests that the increased variance of noise correlations in the wS1 of Vgat-Het mice may reflect weak modulation of the cortical network by behavioral state. Removing noise correlations substantially improved sensory information, consistent with a greater amount of detrimental correlations in Vgat-Het mice. Our findings highlight the altered distribution of noise correlations as a potential circuit endophenotype that could be used to stratify different neurodevelopmental disorders.
We propose that the reduced stimulus-evoked response in inhibitory neurons may be a contributing factor to the increased variance of noise correlations in Vgat-Het mice, based on recent works. Recent studies have used a spiking network model to recapitulate the noise correlations observed in vivo and investigated the influence of different model parameters on these correlations (Stringer et al., 2016). One notable finding was that the strength of noise correlations had an inverse relationship with the magnitude of inhibitory feedback, indicating that strong inhibitory feedback was associated with weaker noise correlations (Stringer et al., 2016). This aligns with experimental data demonstrating that a decrease in noise correlations was accompanied by an increase in activity of fast-spiking interneuron (Stringer et al., 2016). Knocking out a copy of the SYNGAP1 gene in cortical inhibitory interneurons disrupts PV cell-mediated perisomatic innervation of pyramidal neurons, which in turn reduces feedback inhibition (Berryer et al., 2016). Under this condition, even small additions of spikes can be amplified through recurrent excitation, leading to prolonged bursts of activity (up states) (Curto et al., 2009). These up states are followed by strong network adaptation (down states) for an extended period before the cycle repeats (Curto et al., 2009; Mochol et al., 2015; Stringer et al., 2016). These stochastic transitions into silenced periods following bursts of activity are thought to promote synchrony and noise correlations within the cortical network (Mochol et al., 2015). In this regimen, the addition of a small number of spikes through external inputs can induce anticorrelations even among neuron pairs that exhibit correlated activities during spontaneous conditions (Stringer et al., 2016). This further contributes to an increased variance of noise correlations. Conversely, when the feedback inhibition is intact, it suppresses burst activity, and the transitions between up and down states become less pronounced. Consequently, externally added spikes do not elicit strong correlations or anticorrelations among neuronal activities in this regimen.
While we report cortical circuit disruptions associated with specific observed behavioral deficits, we do not claim that wS1 is the only brain structure contributing to the deficits described here. SYNGAP1 is most abundantly expressed in the cerebral cortex and hippocampus during brain development, but a modest level of expression is also detected in the striatum (Komiyama et al., 2002; Araki et al., 2020). Although Vgat is expressed in both GABAergic and glycinergic neurons (Wang et al., 2009), glycinergic neurons are sparse in the neocortex where SYNGAP1 is abundant. Therefore, our intersectional genetic approach predominantly targets SYNGAP1 expression in GABAergic inhibitory interneurons in the cortex and hippocampus. The behavioral deficits reported here are likely to originate from circuit disruptions in these areas. Future studies could examine the role of SYNGAP1 in the striatum and associated behaviors, using circuit-specific manipulations. Another potential caveat in interpreting our results is that SYNGAP1 haploinsufficiency may disrupt the ascending sensory processing pathway. This is unlikely, however, as SYNGAP1 haploinsufficiency has little effect on the development of the somatosensory barrel map or thalamocortical innervation in wS1 (Barnett et al., 2006).
Whether circuit abnormalities observed in the wS1 causally drive the observed behavioral deficits of Vgat-Het mice, or whether other cortical regions are also involved in this process, are important remaining questions. SYNGAP1 haploinsufficiency exerts differential effects on PV neurons — a major cortical GABAergic interneuron subtype — in different cortical regions. PV neurons are reduced in number in the PFC but not in wS1 of pan-neuronal SYNGAP1 heterozygous KO mice (Sullivan et al., 2020). Interestingly, in these mice, expression of the AMPA receptor subunit GluA2 is selectively elevated in PV neurons that are located in wS1 but not in those of the PFC (Sullivan et al., 2020). SYNGAP1 expression in PV cells is known to be important for perisomatic innervation of excitatory pyramidal neurons in wS1 (Berryer et al., 2016). SYNGAP1 is also expressed in SST and vasointestinal peptide inhibitory interneurons (Fig. 1A). Neuronal subtype-specific functions of SynGAP and how these are disrupted by SYNGAP1 haploinsufficiency warrant further investigation.
Collectively, our results add to growing evidence highlighting the contribution of cortical inhibitory interneurons to sensorimotor abnormalities in neurodevelopmental disorders (Contractor et al., 2021). Furthermore, we identify circuit-level endophenotypes in the primary somatosensory cortex that likely contribute to sensorimotor impairments. These findings have direct implications for studies focusing on circuit dysfunction in autism spectrum disorders.
Footnotes
The authors declare no competing financial interests.
This work was supported by Simons Foundation Autism Research Initiative Bridge to Independence Grant to S.E.K. We thank members of the S.E.K. laboratory, Dr. Sara Aton and Dr. Pamela Raymond for critical reading of the manuscript.
- Correspondence should be addressed to Sung Eun Kwon at sungeun1020{at}gmail.com
