Upregulated NMDAR-mediated GABAergic transmission underlies autistic-like deficits in Htr3a knockout mice

Mutations in serotonin pathway genes, especially the serotonergic receptor subunit gene HTR3A, are associated with autism. However, the association of HTR3A deficiency with autism and the underlying mechanisms remain unknown. Methods: The Htr3a knockout (KO) mice were generated using transcription activator-like effector nuclease technology. Various behavior tests, including social interaction, social approach task, olfactory habituation/dishabituation, self-grooming, novel object recognition, contextual fear conditioning, elevated plus maze, open field and seizure susceptibility, were performed to assess the phenotypes. Transcriptome sequencing was carried out to search for molecular network and pathways underlying the phenotypes. Electrophysiological recordings, immunoblotting, immunofluorescence staining, immunoprecipitation, and quantitative real-time PCR were performed to verify the potential mechanisms. The N-methyl-D-aspartate receptor (NMDAR) antagonist memantine was used to treat the KO mice for rescuing the phenotypes. Results: The Htr3a KO mouse model showed three phenotypic domains: autistic-like behaviors (including impaired social behavior, cognitive deficits, and increased repetitive self-grooming), impaired memory, and attenuated susceptibility to pentylenetetrazol-induced seizures. We observed enhanced action potential-driven γ-aminobutyric acid-ergic (GABAergic) transmission in pyramidal neurons and decreased excitatory/inhibitory (E/I) ratio using the patch-clamp recording. Transcriptome sequencing on the hippocampus revealed the converged pathways of the dysregulated molecular networks underlying three phenotypic domains with upregulation of NMDAR. We speculated that Htr3a KO promotes an increase in GABA release through NMDAR upregulation. The electrophysiological recordings on hippocampal parvalbumin-positive (PV+) interneuron revealed increased NMDAR current and NMDAR-dependent excitability. The NMDAR antagonist memantine could rescue GABAergic transmission in the hippocampus and ameliorate autistic-like behaviors of the KO mice. Conclusion: Our data indicated that upregulation of the NMDAR in PV+ interneurons may play a critical role in regulating GABAergic input to pyramidal neurons and maybe involve in the pathogenesis of autism associated with HTR3A deficiency. Therefore, we suggest that the NMDAR system could be considered potential therapeutic target for autism.

Tables (An excel table containing 19 sheets is uploaded separately)   Table S7. The nodes information list of hippocampal differential interactome network. Table S8. The list of 1,036 ASD candidate genes. Table S9. The list of 647 epilepsy candidate genes.

Behavior tests
1. Animal housing and handling. Homozygous Htr3a knockout mice (Htr3a -/-) and WT littermate controls were generated by breeding heterozygous mice. Mice were group housed (4-5 mice per cage) under Specific Pathogen Free (SPF) conditions, given a 12h:12h light-dark cycle and allowed ad libitum access to food and water. Before behavioral tests, mice were handled for 3 days and taken into the testing rooms 30-60 minutes. 2. Social approach task. The procedure for social approach task was slightly modified from the method described previously [1]. Specifically, the testing apparatus was a rectangular clear Plexiglas three chambers box (60 cm (L) x 40 cm (W) x 20 cm (H)). The dividing walls had doorways allowing mouse access to each chamber. The stranger mice from the same strain were habituated to placement inside the wire cage for 5 days prior to testing. Each test mouse was first placed into the center chamber with open access to both left and right chamber, each chamber containing an empty round wire cage. The wire cage (12 cm (H), 11 cm diameter) allows nose contact between mice but prevents fighting. After 10 min of habituation, during the social phase, an age-matched stranger was placed in the one wire cage while the opposite one is empty. The test mouse was allowed to freely explore the social apparatus for 10 min to test whether it prefers to interact with the object (O) or the stranger mouse (S1). Sniffing time was plotted as a social preference index = TS1/(TS1+TO), TS1 -time for a testing mouse interacting with a novel mouse (S1, Stranger1), TO -time for a testing interacting with an empty cage (O, Object). To evaluate the preference for a novel stranger, the test mouse was then tested in a second 10-min session, which contains a novel stranger (S2) in the opposite wire cage. Sniffing time was plotted as a social preference index = TS2/(TS1+T S2), TS1 -time for a testing mouse interacting with a familiar mouse (S1, Stranger 1), TS2 -time for a testing mouse interacting with a novel mouse (S2, Stranger 2). The duration of sniffing, defined as positioning of the nose of the test mouse within 2.5 cm of a cage, was measured using software EthoVision XT11.5 (Noldus).
3. Home cage social interaction test. The social interaction test was performed as previously described [2]. Each mouse was left alone in its home cage for 15 min. An unfamiliar male 4 C57BL/6N mouse of the same age was then introduced. The behavior of the test mouse was video-recorded for 10 min and scored the time of active interactions, including sniffing, allo-grooming, mounting and following. 4. Olfactory habituation/dishabituation test. This test was conducted as previously described [3].
Each subject mouse was tested in a clean mouse cage. Cotton tipped swabs were used to deliver odor stimuli. Olfactory cues were designed to measure familiar or unfamiliar odors, with or without social odors. Three identical swabs (2-min for each swab) were orderly assayed for the habituation to the same odor. Water, almond odor (prepared from almondretrieve, 1:100 dilution in tap water), banana odor (prepared from imitation banana flavor, 1:100 dilution), odor from cage 1 (social odor 1), odor from cage 2 (social odor 2) were presented in sequence to assay the dishabituation to different odors. Water, almond odor, and banana odor were prepared by dipping the cotton tip in the solution for 2 sec. Social odors were prepared by wiping a swab in pattern across a soiled cage of unfamiliar mice of the same sex. Time spent sniffing the swab was quantitated with a stopwatch by an observer. Sniffing time was scored when the distance between mouse's nose and the swab was 1 cm or shorter. 5. Self-grooming test. Mouse was placed in an empty cage without bedding. After 10-min habituation, mouse behaviors were recorded for another 10-min. Self-grooming behavior was defined as stroking or scratching of the body or face, or licking body parts. The cumulative time spent in grooming all body regions were evaluated by using a stopwatch as described previously [4]. 6. Novel object recognition [5]. Short habituation session, mouse was placed into a Plexiglas rectangular cage (22 cm height × 44 cm length × 22 cm width) for 5 min. In the familiarization session (twenty-four hours after habituation session), the mouse was presented with a pair of identical objects (either towers of Lego bricks or Falcon tissue culture flasks) 5 cm away from the walls. The time when a mouse shows any investigative behaviors (head orientation or sniffing occurring, or entering an area within 1 cm around the object), is considered as exploring time. Stopwatch was used to record the time spent exploring each object until the total exploring time reached 20 seconds. During the testing trial (testing phase, performed 24 hours later), one of the familiar objects was replaced by a novel object. The 5 exploring time for the familiar or the novel object during the test phase was recorded until 20 seconds of total exploring time was reached. 7. Contextual fear conditioning. During training, mouse was first allowed to freely explore the apparatus (MED-VFC-NIR-M; Med Associates) for 3 min, and then exposed to 4 times of tone-foot shock pairings (tone, 30 sec, 80 dB; foot shock, 1 sec, 0.75 mA) with an interval of 80 sec. Twenty four hours after training, mouse was returned to the chamber for 2 minutes to evaluate contextual fear memory. The percentage of freezing time during training and testing was measured using Med Associates Video-Tracking and scoring software.  4) clonic (tonic seizure, tonic hindlimb extension, or death). 6 11. Memantine treatment. Memantine was purchased from SIGMA (USA) and dissolved in saline.
Memantine treatment was conducted as previously described [7]. Memantine (5 mg/kg) or saline alone (control) was administered to mice by intraperitoneal (i.p.) injection 30 min before behavior tests, such as the social interaction test, social approach task, measurements of repetitive self-grooming behaviors or the injection of PTZ. For behavior tests that require training session and test session, such as novel object recognition test and contextual fear conditioning test, memantine (5 mg/kg) or saline was i.p. injected into mice 30 min before the training session and 30 min before test session. Behavior tests were performed as described above. Electrophysiological recordings were performed as described above, with/without the bath containing 1 µM memantine. The current changes were evaluated before, and 5 min after application of memantine.

RNA-seq and differential expression analysis
Each RNA sample was extracted from dissected hippocampi of adult mice according to the manufacturer's protocol (RNAeasy Mini Kit, Qiagen, USA). The quality and yield of the isolated RNAs were assessed using a NanoDrop Spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA) and Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Only RNAs with a high RNA integrity number (RIN > 9) were selected and used for the subsequent sequencing. RNA sequencing was performed at Novogene (Beijing, China) using Illumina NovaSeq. The paired-end reads were aligned to the reference mice genome (mm10 assembly) and low-quality regions were removed. A Picard tool, MarkDuplicates, was used to mark duplicate reads. Reference genome (mm10) and annotation files were downloaded from UCSC Genome Browser. Reads numbers mapped to each gene were counted using HTseq-count (v0.9.0) [8]. Genes with counts > 4 counts in at least 4 of 6 samples were defined as expressed genes in the analysis with DESeq2 (v1.20.0) [9]. Genes with CPM (count-per-million) >1 in at least one of six samples were considered as expressed genes in the analysis with edgeR (v3.22.5) [10]. Hippocampal expressed genes are intersection of the expressed genes identified by DESeq2 and edgeR. Heatmap of differentially expressed genes (DEGs) and principal component analysis (PCA) were carried out based on regularized log2-transformed data using pheatmap package and plotPCA in DESeq2. Differential expression analysis on two groups was performed using the DESeq2 and the edgeR (v3.22.5) [10], using a cutoff of FDR < 0.05 for DESeq2 7 and p < 0.01 for edgeR. Differentially expressed genes (DEGs) was the intersection part of DESeq2 (adj.p < 0.05) and edgeR (p < 0.01).

Functional enrichment analyses for differentially expressed genes
1. Functional annotations of up-regulated and down-regulated genes were done using Database for Annotation, Visualization and Integrated Discovery (DAVID) [11] tools (v6.8) and terms were identified with FDR less than 0.05. Visualization and plot of top selected terms were done using ggplot2 package (v3.0.0).
Human gene names were converted to orthologous mouse genes using the Ensemble BioMart [16]. The p-values of enrichment of disease-related genes in DEGs was calculated using Fisher's Exact Test. The same method was used for up-regulated genes and down-regulated genes.

Construction of hippocampal interactome and DEG Network
We constructed a hippocampal interactome by mapping 16,435 expressed genes from the mouse hippocampal transcriptome to the whole mouse interactome from BioGRID [17], which contains 4,353 nodes and 9,618 edges. We mapped the 2,092 DEGs to the mouse hippocampal interactome to retrieve hippocampal DEG Network containing DEGs and their first co-expressed neighbors. Co-expression relationship was determined by a cutoff 0.75 of correlation coefficient, which was calculated based on FPKM value (fragments per kilobase of transcript per million mapped reads) using WGCNA [18].
Self-loop edges and zero-degree nodes were removed.

Networks for autism, epilepsy and learning/memory
We mapped the 1,036 ASD candidate genes to the mouse hippocampal interactome to retrieve an ASD Network containing ASD candidate genes and their first co-expressed neighbors. The same method was also applied for retrieving epilepsy Network and learning/memory Network.

(C) Sociability (left bar plot):
In the 10-min sociability phase of the social approach task, both Htr3a -/and WT mice showed preference to interact with a stranger mouse (S1, Stranger1) rather than an empty cage (O, Object) (two-way ANOVA test, p = 0.0002, n = 12 for WT mice, p = 0.0090, n = 11 for Htr3a -/mice). Social novelty (right bar plot): In the 10-min social novelty phase of the social approach task, Htr3a -/mice showed no significant preference to interact with a stranger mouse (S2, Stranger 2) over a familiar mouse (S1, Stranger 1), while WT mice showed preference to interact with the stranger mouse (S2) over the familiar mouse (S1) (two-way ANOVA test, p = 0.0017, n = 12 for WT mice; p = 0.9308, n = 11 for Htr3a -/mice).

Figure S2. Female Htr3a KO mice exhibited impaired social behavior and memory.
(A) In the 10-min sociability phase of the social approach task, there was no significant difference between female knockout and WT mice (S1, Stranger1; O, Object) (Student's t test, p = 0.2426).
(B) Compared with WT mice, female Htr3a -/mice showed significantly decreased preference to interact with a stranger mouse (S2, Stranger 2) over a familiar mouse (S1, Stranger 1) in the 10-min social novelty phase of the social approach task (Student's t test, p = 0.0198). After 24 hours, female knockout mice showed less freezing time in the contextual fear memory test than WT mice (Student's t test, p = 0.0035).
(E) The protein interaction network for DEGs (DEG Network) consists of 245 nodes and 222 edges. Dotted circles indicate the subnetworks (the major components of the network). Each of these subnetworks is enriched with indicated function, and is thus considered to be a functional module (marked as M1-6).
Red node: upregulated; blue node: downregulated; gray node: without expression change; node with green border: co-expressed neighbor; gray line: protein-protein interaction (PPI); double lines: PPI and co-expression. Figure S5. Comparison of the enriched pathways between ASD-, EP-and LM Networks. 15 There were 47 enriched pathways shared by three networks. The enrichment analysis was performed using DAVID functional annotation tool with an adjusted p-value cutoff 0.05.