Mice lacking exon 3 of the Slc6a20a gene were previously described (Bae et al., 2021). All mice were maintained under a 12-hr light/dark cycle (light phase: between 1:00 a.m. and 1:00 p.m.), given food and water ad libitum, and weaned at postnatal day 21. Female heterozygous mice and male heterozygous mice were crossed in order to obtain wild-type, heterozygous, and homozygous progenies. All mice were maintained in the mouse facility of Korea Advanced Institute of Science and Technology (KAIST). The experimental procedures were approved by the Committee of Animal Research at KAIST (KA2020-89).

Adult male mice aged 3-6 months were used for all behavioral tests. All behavioral experiments were performed during the dark phase of the light/dark cycle (1:00 p.m. to 1:00 a.m.) in which mice are usually active. The brightness conditions in behavioral experiments described below refer to the local brightness around the center region of the apparatus rather than that in the behavioral room. Mice were habituated to an empty, dark experimental room for 30 min before the start of each behavioral test except for the Morris water maze test. EthoVision XT (Noldus) was used to analyze behavioral results unless otherwise noted.

This experiment was performed in order to analyze the locomotor activity and anxiety-like behavior in mice. A mouse was placed into the center of a white box (40 cm x 40 cm x 40 cm; ∼100 lux in the center). Mouse movements were then recorded for an hour, and the distance moved and time spent in the center region of the apparatus were analyzed.

This experiment was performed to investigate height-induced anxiety-like behavior in mice. A heightened, cross-shaped device with two arms closed and two arms open was used for the experiment. A mouse was placed in the center region of the apparatus (∼200 lux) and allowed to freely move around the environment for 10 minutes. Time spent in closed arms, open arms, and center was analyzed.

This experiment was performed to measure light-induced anxiety-like behavior in mice. The apparatus contained two different boxes/chambers, with one with the roof (dark box), and another without the roof (light box; ∼250 lux). A mouse was placed in the light box at the beginning and allowed to freely move around the environment.

This test was performed to measure social approach and social novelty recognition (Crawley, 2004; Nadler et al., 2004). A mouse was placed in the center chamber of the three-chambered apparatus (59 cm x 39.5 cm x 21.5 cm; ∼100 lux), and its activity was recorded for 10 min. Then, a stranger mouse (S1) was positioned in the cage placed at the corner of a side chamber, and an object (O) was placed in the cage at the corner of the other side chamber. 129S1 mice from the Jackson Laboratory were used as strangers. After 10-min recording of mouse activity, the object was replaced with another stranger (S2), followed by 10-min recording.

This test was performed to measure spatial learning and memory in mice (Vorhees and Williams, 2006). A round pool was filled with white paint-added water. Water temperature was maintained at ∼20°C. The pool was divided into four quadrants, and a platform was positioned in one of the quadrants. For the first six days, mice were taught to find an invisible platform. On the seventh day, the probe test was performed: a platform was removed from the pool. A mouse was then placed in the center of the pool and allowed to freely swim around for a minute. Time spent in quadrants and number of crossings across the platform were measured. After the probe test, a platform was placed in the opposite quadrant. For additional three days, mice were trained to locate the new position of the platform (reversal phase). On the eleventh day, another probe test (reversal probe test) was performed.

This test was performed to measure fear learning and memory in mice. On day 0, a mouse was placed in the chamber with an electrocuting platform and habituated for 5 min. On day 1, foot shocks (0.8 mA) were provided for 2 s at 2, 3, and 4 min after the start of recording. Activity of a mouse was recorded for 5 min. On days 2, 3, and 8, a mouse was placed into the same chamber without any shock, and its activity was recorded for 5 min. Percent of freezing time was calculated.

LABORAS (Laboratory Animal Behavior Observation Registration and Analysis System) experiments were performed for a long-term (4-day) monitoring of various mouse behaviors such as locomotor activity, repetitive behaviors (climbing, grooming, and rearing), drinking, and eating (Quinn et al., 2003, 2006). Total investigation time was 96 h with 12-h light/dark cycle, during which food and water were provided.

Four mice at ∼P120 were used for each group (wild type, heterozygous, and homozygous). The extracted mouse brains were preserved in RNAlater solution (Ambion) and stored at −20°C. Poly-T oligo-attached magnetic beads were utilized to purify poly-A mRNAs. RNA concentrations were quantified using Quant-IT RiboGreen (Invitrogen, R11490), and RNA integrity was determined using TapeStation RNA screen tape (Agilent Technologies), after which only high-quality RNAs (RIN > 7.0) were selected for cDNA library construction using Illumina TruSeq mRNA Sample Prep kit (Illumina). Indexed libraries were submitted to an Illumina NovaSeq (Illumina), and paired-end (2 × 100 bp) sequencing was performed by Macrogen Inc.

Transcript abundance was estimated with Salmon (v 1.1.0) (Patro et al., 2017) in Quasi-mapping-based mode onto the Mus musculus genome (GRCm38) with GC bias correction (−gcBias). The acquired abundance data was imported to R (v.3.5.3) with tximport (Soneson et al., 2015) package and differential gene expression analysis was performed using R/Bioconductor DEseq2 (v1.30.1) (Love et al., 2014). Principal component analysis (PCA) was performed for the regularized log transform (rlog) of the normalized counts using plotPCA (with default parameter) tools implemented in DEseq2. Normalized read counts were computed by dividing the raw read counts by size factors and fitted to a negative binomial distribution. The p-values were adjusted for multiple testing with the Benjamini-Hochberg correction. Genes with an adjusted p-value of less than 0.05 were considered as differentially expressed. Volcano plots were generated using R ggplot2 (v.3.1.1) package. The Gene Ontology (GO) enrichment analyses were performed using DAVID software (version 6.8) (Huang da et al., 2009). Mouse gene names were converted to human homologs using the Mouse Genome Informatics (MGI) database. Gene Set Enrichment Analysis (GSEA) (Subramanian et al., 2005, 2007) was performed to determine whether a priori-defined gene sets would show statistically significant differences in expression between WT and Slc6a20a-mutant mice. Enrichment Analysis was performed using GSEAPreranked^® (gsea-3.0.jar) module on gene set collections downloaded from Molecular Signature Database (MSigDB) v 7.0. GSEAPreranked was applied using the list of all genes expressed, ranked by the fold change and multiplied by the inverse of the p-value with recommended default settings (1,000 permutations and a classic scoring scheme). The false discovery rate (FDR) was estimated to control the false positive finding of a given normalized enrichment score (NES) by comparing the tails of the observed and null distributions derived from 1,000 gene set permutations. The gene sets with an FDR of less than 0.05 were considered as significantly enriched. Integration and visualization of the GSEA results were performed using the EnrichmentMap Cytoscape App (version 3.8.1) (Merico et al., 2010; Isserlin et al., 2014).

GraphPad Prism (version 9.2.0; GraphPad Software) were used to perform the statistical analyses. Outliers were retained. Statistical details are presented in Supplementary Table 1.

Because glycine levels and NMDAR functions are elevated in Slc6a20a-mutant mice (Bae et al., 2021), and NMDARs critically regulate brain development and function (Paoletti et al., 2013), we subjected Slc6a20a heterozygous and homozygous mutant mice (Slc6a20a^+/– and Slc6a20a^–/– mice, respectively) to a battery of behavioral tests.

As the first step, we used immunoblot analysis to determine the temporal pattern of Slc6a20a protein expression during brain development. The immunoblot pattern from whole-brain lysates indicated that protein expression was strong at early developmental stages, including late embryonic and neonatal stages, and thereafter decreased to adult levels over the first few weeks of postnatal life (Figure 1A).

In the open-field test, Slc6a20a^+/– and Slc6a20a^–/– mice (3–6 months) showed locomotor activities comparable to those of wild-type (WT) mice (Figure 1B), suggestive of normal locomotor activity in a novel environment. In the Laboras test, where mouse movements were monitored for 4 consecutive days (Quinn et al., 2003, 2006), Slc6a20a^+/– mice showed normal locomotor activity comparable to that of WT mice (Figure 1C). In contrast, Slc6a20a^–/– mice showed moderately increased locomotor activity during the total period and the light-on period, but not during the light-off period (Figure 1C), suggesting that homozygous but not heterozygous Slc6a20a deletion in mice leads to hyperactivity in a familiar environment. The increased activity during the light-on period may also suggest that sleep was disturbed in Slc6a20a^–/– mice, as reported previously in mice lacking the receptor tyrosine phosphatase PTPRS and showing disturbed sleep behaviors and rhythms (Park et al., 2020).

Slc6a20a^+/– and Slc6a20a^–/– mice spent normal amounts of time in the center region of the open-field arena (Figure 1B), suggesting that anxiety-like behavior was not altered. Similarly, Slc6a20a^+/– and Slc6a20a^–/– mice showed normal levels of anxiety-like behaviors in the elevated plus-maze and light-dark tests (Figures 1D,E).

These results collectively suggest that homozygous but not heterozygous Slc6a20a deletion in mice induces moderate hyperactivity in a familiar environment, without altering anxiety-like behavior.

We next subjected Slc6a20a-mutant mice to behavioral tests in the social and repetitive behavioral domains. In the three-chamber social-interaction test, Slc6a20a^+/– and Slc6a20a^–/– mice displayed normal levels of social approach, as shown by time spent exploring social and non-social targets and the social preference index (Figure 2A). In addition, these mice displayed normal levels of social novelty recognition, as shown by time spent exploring novel and familiar social targets (Figure 2A).

In the Laboras test, Slc6a20a^–/– but not Slc6a20a^+/– mice showed increased repetitive climbing (Figure 2B), a form of repetitive behavior characterized by overhanging from the wire cage lids (Protais et al., 1976; Riffee et al., 1979; Wilcox et al., 1979; Cabib and Puglisi-Allegra, 1985). However, Slc6a20a^+/– and Slc6a20a^–/– mice did not show any alteration in other repetitive behaviors, such as self-grooming and rearing (Figure 2B). Other behaviors, such as drinking and eating, were also normal in Slc6a20a^–/– and Slc6a20a^+/– mice (Supplementary Figure 1).

These results suggest that Slc6a20a deficiency in mice induces a moderate increase in repetitive behavior without affecting social behaviors, with repetitive climbing but not other repetitive behaviors increased in Slc6a20a^–/– but not Slc6a20a^+/– mice.

Because NMDARs critically regulate various forms of learning and memory (Collingridge, 1987; Bliss et al., 2014), we next subjected Slc6a20a^+/– and Slc6a20a^–/– mice to spatial and fear learning and memory paradigms.

In the Morris water maze test, Slc6a20a^+/– and Slc6a20a^–/– mice showed normal levels of spatial memory acquisition and retrieval during the initial learning and probe phases of the test, respectively, compared with WT mice (Figure 3A). In the reversal test, Slc6a20a^+/– but not Slc6a20a^–/– mice showed enhancements during the acquisition but not probe phase, compared with WT mice (Figure 3A).

In the contextual fear memory conditioning test, Slc6a20a^+/– mice showed normal levels of fear memory acquisition on day 1 compared with WT mice (Figure 3B). In memory retrieval tests consecutively performed on day 2 (for 24-h retrieval), day 3 (for 48-h retrieval), and day 8 (7-day retrieval), Slc6a20a^+/– mice showed largely normal fear memory retrieval.

Slc6a20a^–/– mice showed normal levels of fear memory acquisition on day 1 (Figure 3B), although there was an increasing tendency. In the retrieval tests, Slc6a20a^–/– mice showed increased retrieval on day 3 (48-h retrieval); the levels of retrieval on day 2 (24-h retrieval) and day 8 (7-day retrieval) were normal but showed increasing tendencies.

These results collectively suggest that Slc6a20a deficiency in mice induces moderate increases in spatial learning and memory in the Morris water maze test and fear learning and memory in contextual fear conditioning test.

To investigate the molecular phenotypes of Slc6a20a^+/– and Slc6a20a^–/– mice, we performed RNA-sequencing (RNA-seq) analyses using whole-brain lysates obtained at ∼P120. Our analysis identified 13 DEGs (7 upregulated and 6 downregulated; cutoff, p value < 0.05) in Slc6a20a^+/– mice and 33 DEGs (27 upregulated and 6 downregulated) in Slc6a20a^–/– mice, with four DEGs overlapping between the genotypes (Nlgn3/neuroligin-3, Tenm3/teneurin-3, and Cdc73 upregualted and Slc6a20a downregulated) (Figures 4A,B and Supplementary Table 2).

In DEGs from Slc6a20a^+/– mice, the seven significantly upregulated genes were Cdc73 (cell division cycle 73), Nlgn3 (neuroligin 3), Camsap1 (calmodulin regulated spectrin-associated protein 1), Ypel2 (Yippee like 2), Shisa9 (Shisa family member 9), Ppp1r26 (protein phosphatase 1 regulatory subunit 26), and Tenm3 (teneurin transmembrane protein 3) (Figure 4C and Supplementary Table 3). The products of some of these genes have been associated with synaptic and neuronal functions. Neuroligin-3 is a synaptic adhesion molecule involved in synapse development and regulation and ASD-related brain functions and behaviors (Etherton et al., 2011; Foldy et al., 2013; Jaramillo et al., 2014; Rothwell et al., 2014; Zhang and Sudhof, 2016; Cao et al., 2018; Sudhof, 2018). CAMSAP1 binds to the minus end of microtubules and regulates neuronal polarity/migration and cortical lamination (Meng et al., 2008; Akhmanova and Steinmetz, 2015; Zhou et al., 2020). Shisa9 (also known as CKAMP44) is an AMPA receptor auxiliary protein that regulates the trafficking, subcellular localization, and function of AMPA receptors (von Engelhardt et al., 2010; Jacobi and von Engelhardt, 2018; von Engelhardt, 2019). Teneurin-3 is a transmembrane protein involved in homophilic adhesion as well as heterophilic adhesion with latrophilins that regulates synapse specificity and neural circuit assembly (Mosca et al., 2012; Li et al., 2018; Sando et al., 2019; Pederick and Luo, 2021).

The six significantly downregulated genes in Slc6a20a^+/– mice were Slc6a20a (solute carrier family 6 member 20), as expected, followed by Col24a1 (collagen type XXIV alpha 1 chain), Nacad (NAC alpha domain-containing), Alas2 (5′-aminolevulinate synthase 2), and Serhl2 (serine hydrolase-like 2), and Snrnp70 (small nuclear ribonucleoprotein U1 subunit 70) (Figure 4C). Interestingly, Alas2 is an enzyme localized in the mitochondria of erythrocytes; it regulates the heme biosynthetic pathway and is implicated in X-linked sideroblastic anemia (XLSA) (Nzelu et al., 2021).

In Slc6a20a^–/– mice, the top ten upregulated genes were Gtf3c4 (general transcription factor IIIC subunit 4), Dlx1 (distal-less homeobox 1), Gng4 (G protein subunit gamma 4), Cdc73 (cell division cycle 73; also upregulated in Slc6a20a^+/– mice), Atp1b2 (ATPase Na + /K + transporting subunit beta 2), Robo1 (roundabout guidance receptor 1), Tenm3 (teneurin transmembrane protein 3), Pcdhga4 (protocadherin gamma subfamily A, 4), Cdhr1 (cadherin related family member 1), and Eomes (eomesodermin) (Figure 4D). Dlx1 is a homeobox transcription factor that regulates neuronal differentiation (Petryniak et al., 2007; Lee B. et al., 2018; Pla et al., 2018; Lindtner et al., 2019). Atp1b2 is a sodium-potassium ATPase that regulates neuronal excitability (Larsen et al., 2016). Robo1 is a transmembrane protein that regulates axon guidance and neuronal migration (Seeger et al., 1993; Kidd et al., 1998).

Many DEGs from Slc6a20a^–/– mice not mentioned above have also been associated with synaptic and neuronal functions; examples include Mdga1 (MAM domain containing glycosylphosphatidylinositol anchor 1) (Lee et al., 2013; Pettem et al., 2013; Kim et al., 2017; Um and Ko, 2017), Cacng8 (calcium voltage-gated channel auxiliy subunit gamma 8) (Nicoll et al., 2006; Diaz-Alonso and Nicoll, 2021), Sema5a (semaphorin 5A) (Duan et al., 2014), and Nlgn3 (neuroligin 3) (Sudhof, 2018). It is also notable that neuroligin-3 and teneurin-3 are similarly upregulated in Slc6a20a^+/– and Slc6a20a^–/– mice, suggesting that their upregulations may represent a shared mechanism responding to Slc6a20a deletion. Lastly, the six significantly downregulated genes in Slc6a20a^–/– mice were Slc6a20a, Rpl9 (ribosomal protein L9), Col11a1 (collagen type XI alpha 1 chain), Tubgcp6 (tubulin gamma complex associated protein 6), Rpl13a (ribosomal protein L13A), and Sub1 (SUB1 homolog, transcriptional regulator) (Figure 4D).

These results collectively suggest that both Slc6a20a^+/– and Slc6a20a^–/– mice show upregulation of genes that are associated with synaptic and neuronal functions and downregulation of genes associated with ribosomal and mitochondrial functions. Similar results were obtained from our gene set enrichment analysis (GSEA; see below).

Since our DEG analysis yielded small numbers of DEGs, we performed gene set enrichment analysis (GSEA), which can identify altered biological functions using a large number of small but coordinated transcriptomic changes in a less biased manner than analysis of a small number of large transcriptomic changes (Subramanian et al., 2005, 2007).

The transcripts derived from WT and Slc6a20a^+/– mice (HT/WT transcripts; ∼P120 whole brain) were positively enriched for gene sets associated with neuronal synapses, as shown by the top five enriched gene sets (Figure 5A and Supplementary Table 4). Clustering of the positively enriched gene sets using CytoScape Enrichment App (Merico et al., 2010; Isserlin et al., 2014) further confirmed that there were positive enrichments for functions associated with neuronal synapses, such as synaptic specialization, presynaptic active zone, ion channels, dendrites, and axons (Figure 5B).

These results were obtained using the gene sets of the cellular component domain in the C5 database; we also obtained similar results using the gene sets of the biological process and molecular function domains of the C5 database (Supplementary Figures 2A,B, 3A,B and Supplementary Table 4). In addition, the use of gene sets in the KEGG domain indicated positive enrichments for synapse-related gene sets including long-term potentiation and gap junction, known to cooperate with excitatory synapses (Lee et al., 2021), as well as negative enrichments for ribosome/mitochondria-related gene sets (Supplementary Figures 4A,B and Supplementary Table 4).

The HT/WT transcripts were negatively enriched for gene sets associated with ribosomes and mitochondria (Figure 5A and Supplementary Table 4). In addition, CytoScape Enrichment App analysis revealed similar negative enrichments for ribosome/mitochondria-related functions, such as ribosomal subunits and respiratory chain complex (Figure 5B).

The transcripts derived from WT and Slc6a20a^–/– mice (HM/WT transcripts) were positively enriched for gene sets associated with neuronal synapses, as supported by the top five gene sets and CytoScape Enrichment App clustering (Figures 5C,D and Supplementary Table 4). In addition, the HM/WT transcripts were negatively enriched for gene sets associated with ribosomes and mitochondria, as supported by the top five gene sets and CytoScape Enrichment App clustering (Figures 5C,D and Supplementary Table 4). Similar results were obtained using the gene sets of the biological process and molecular function domains (Supplementary Figures 2C,D, 3C,D and Supplementary Table 4).

These results collectively suggest that heterozygous and homozygous deletion of Slc6a20a in mice leads to similar increases in synapse-associated genes and similar decreases in ribosome- and mitochondria-related genes.

Previous studies investigated transcriptomic changes associated with ASD (Garbett et al., 2008; Voineagu et al., 2011; Gupta et al., 2014; Parikshak et al., 2016; Velmeshev et al., 2019) and reported gene sets that are up- or downregulated in ASD (termed ASD-related gene sets hereafter), including DEG_Up_Voineagu, Co-Exp_Up_M16_Voineagu, DEG_Down_Voineagu, and Co-Exp_Down_M12_Voineagu (Voineagu et al., 2011; Werling et al., 2016) (details on these gene sets are summarized in Supplementary Table 5).

In addition, a large number of previous studies on ASD led to the compilation of ASD-risk gene sets, including SFARI genes (all genes and high-confidence category 1 genes) (Abrahams et al., 2013)¹, FMRP targets (Darnell et al., 2011; Werling et al., 2016), De Novo Missense (protein-disrupting or missense rare de novo variants) (Iossifov et al., 2014; Werling et al., 2016), De Novo Variants (protein-disrupting rare de novo variants) (Iossifov et al., 2014; Werling et al., 2016), and AutismKB (Autism KnowledgeBase) (Xu et al., 2012; Yang et al., 2018; Supplementary Table 5). The genes in these ASD-risk gene sets are thought be generally downregulated in ASD because many of the mutations are missense, nonsense, splice-site, frame-shift, and deletion mutations.

Using these gene sets, we performed GSEA using the HT/WT and HM/WT transcripts from Slc6a20a-mutant mice. The HT/WT and HM/WT transcripts were positively and similarly enriched for ASD-related gene sets that are downregulated in ASD, such as DEG_Down_Voineagu and Co-Exp_Down_M12_Voineagu (Figure 6A and Supplementary Table 4), suggesting that both HT/WT and HM/WT transcripts exhibit patterns opposite those observed in ASD (termed anti-ASD hereafter).

The HT/WT and HM/WT transcripts were also positively and similarly enriched for all of the tested ASD-risk gene sets, including SFARI genes, FMRP targets (most strongly enriched), De Novo Missense, De Novo Variants, and AutismKB (Figure 6A and Supplementary Table 4), and thus again exhibited an anti-ASD pattern.

When tested against gene sets associated with various brain disorders (DisGeNet² (Pinero et al., 2017, 2020), the HT/WT and HM/WT transcripts were positively and similarly enriched for ASD and epilepsy-related gene sets, and the HM/WT transcripts were positively and strongly enriched for substance use-related gene sets relative to the HT/WT transcripts (Figure 6B and Supplementary Table 4).

Distinct cell-type-specific transcriptomic changes have also been reported in ASD, including downregulation of neuron- and oligodendrocyte-related genes and upregulation of astrocyte- and microglia-related genes (Voineagu et al., 2011; Werling et al., 2016). This led us to test if the HT/WT and HM/WT transcripts are enriched in these previously reported cell type-specific gene sets (Albright and Gonzalez-Scarano, 2004; Cahoy et al., 2008; Kang et al., 2011; Zeisel et al., 2015; Werling et al., 2016; Velmeshev et al., 2019, 2020; Supplementary Table 5).

The HT/WT and HM/WT transcripts were positively enriched for gene sets associated with glutamate and GABA neurons (Figure 6C and Supplementary Table 4), suggestive of anti-ASD transcriptomic patterns. However, the HT/WT and HM/WT transcripts were negatively and moderately enriched for gene sets associated with oligodendrocytes (Figure 6C), suggestive of “ASD-like” or “pro-ASD” transcriptomic patterns. Intriguingly, the HT/WT and HM/WT transcripts displayed weak enrichments for astrocyte and microglia-related gene sets relative to those for glutamate/GABA neurons; thus we observed both positive and negative enrichments for microglia (Figure 6C).

These results collectively suggest that heterozygous and homozygous Slc6a20a deletions in mice lead to largely anti-ASD transcriptomic enrichment patterns, as supported by the enrichment patterns for ASD-related/risk and cell-type-specific gene sets. In addition, the cell-type-specific transcriptomic changes induced by Slc6a20a deletion are stronger in neurons and oligodendrocytes relative to glial cells.