Neuronal network dysfunction in a human model for Kleefstra syndrome mediated by enhanced NMDAR signaling

Epigenetic regulation of gene transcription plays a critical role in neural network development and in the etiology of Intellectual Disability (ID) and Autism Spectrum Disorder (ASD). However, little is known about the mechanisms by which epigenetic dysregulation leads to neural network defects. Kleefstra syndrome (KS), caused by mutation in the histone methyltransferase EHMT1, is a neurodevelopmental disorder with the clinical features of both ID and ASD. To study the impact of decreased EHMT1 function in human cells, we generated excitatory cortical neurons from induced pluripotent stem (iPS) cells derived from KS patients. In addition, we created an isogenic set by genetically editing healthy iPS cells. Characterization of the neurons at the single-cell and neuronal network level revealed consistent discriminative properties that distinguished EHMT1-mutant from wildtype neurons. Mutant neuronal networks exhibited network bursting with a reduced rate, longer duration, and increased temporal irregularity compared to control networks. We show that these changes were mediated by the upregulation of the NMDA receptor (NMDAR) subunit 1 and correlate with reduced deposition of the repressive H3K9me2 mark, the catalytic product of EHMT1, at the GRIN1 promoter. Furthermore, we show that EHMT1 deficiency in mice leads to similar neuronal network impairments and increased NMDAR function. Finally, we could rescue the KS patient-derived neuronal network phenotypes by pharmacological inhibition of NMDARs. Together, our results demonstrate a direct link between EHMT1 deficiency in human neurons and NMDAR hyperfunction, providing the basis for a more targeted therapeutic approach to treating KS.


Introduction
Advances in human genetics over the past decade have resulted in the identification of hundreds of genes associated with intellectual disability (ID) and autism spectrum disorder (ASD) 1 . Within this group of newly identified genes the number of chromatin regulators is remarkably high [2][3][4] . These regulators are engaged in genome-wide covalent DNA modifications, post-translational modifications of histones, and control of genomic architecture and accessibility 5 . Several ASD/IDlinked chromatin regulators directly interact with one another to form larger complexes, regulating chromatin structure to control the expression of genes important for neurodevelopment and/or neuroplasticity 3 . Despite considerable progress in elucidating the genetic architecture of many neurodevelopmental disorders (NDDs), a large gap still exists between the genetic findings and deciphering the cellular or molecular pathobiology 6 . In particular, we require a better understanding of the relevance of genetic changes with respect to their downstream functional consequences and whether there is overlap between different patients within the clinical spectrum 6 .
Kleefstra syndrome (KS) (OMIM#610253) is an example of a rare monogenic NDD with ID, ASD, hypotonia and dysmorphic features as hallmark phenotypes [7][8][9] . KS is caused by heterozygous de novo loss-of-function mutations in the EHMT1 gene (Euchromatin Histone Lysine Methyltransferase 1, also known as GLP) or by small 9q34 deletions harboring (part of) the gene 7 .
Constitutive and conditional loss of EHMT1 function in mice and Drosophila leads to learning and memory impairments [11][12][13] . Additionally, Ehmt1 +/mice recapitulate the developmental delay and autistic-like behaviors that are observed in KS patients 14,15 . At the cellular level, these mice show a significant reduction in dendritic arborization and the number of mature spines in CA1 pyramidal neurons 11 . The dynamic regulation of H3K9me2 by EHMT1/2 is also involved in synaptic plasticity and learning and memory [16][17][18] . For instance, EHMT1 and 2 are required for synaptic scaling, a specific form of homeostatic plasticity, by regulating the expression of brain-derived neurotrophic

Characterization of KS patient-derived neurons at the single-cell level
We generated several iPS cell lines, including from two patients with KS and two healthy subjects (respectively KS 1 , KS 2 and C 1 , C 2 ) ( Fig. 1A and Supplementary Fig. S1, Material and Methods).
One patient (KS 1 ) had a frameshift mutation in EHMT1 leading to a premature stop codon  Fig. S2A). In addition to these lines, iPS cells were generated from a parent who had an affected child with KS due to a microdeletion on chromosome 9q34 (233 kb) including EHMT1 that appeared present as a mosaic in the parent 23 Fig. S2E).
Twenty-one days after the start of differentiation (days in vitro, DIV), both control and KS iNeurons showed mature neuronal morphology. We measured this by reconstruction and quantitative morphometry of DsRed-labeled iNeurons at DIV 21. We observed no significant differences between control and KS iNeurons in any aspect of neuronal somatodendritic morphology, including the number of primary dendrites, dendritic length and overall complexity ( Fig. 1D and Supplementary Fig. S3A-H).
ID and ASD have been associated with synaptic deficits in rodents and humans 21,26 . We therefore investigated whether EHMT1-deficiency leads to impairments in synapse formation. To this end, we stained control and KS iNeurons for pre-and post-synaptic markers (i.e. synapsin and PSD95, respectively) at DIV 21. We observed that putative functional synapses were formed on control as well as KS iNeurons (i.e., synapsin always co-localized with PSD95, Supplementary Fig. S3I), without any indications for differences in the number of synaptic puncta between the different iPS cell lines (Fig. 1E). Furthermore, whole-cell patch-clamp recordings at DIV 21 of iNeurons grown at similar densities and in the presence of tetrodotoxin (TTX) also revealed no differences in the frequency or amplitude of AMPA receptor (AMPAR)-mediated miniature excitatory postsynaptic currents (mEPSCs). This indicates that KS iNeurons are generally not impaired in the AMPAR component of the excitatory input they receive ( Fig. 1F and Supplementary Fig. S3J). At DIV 21 nearly 90% of the control and KS patient iNeurons fired multiple action potentials, indicative of a mature state of their electrophysiological properties ( Supplementary Fig. S3K). When recording intrinsic passive and active properties from control and KS patient iNeurons at DIV 21, we found no differences between the lines (Supplementary Fig. S3L-O). Collectively, our data thus indicate that there were no significant differences between control and KS patient iNeurons with regard to neuronal morphology, excitatory synapses and intrinsic properties.

Neuronal networks of KS iNeurons show an aberrant pattern of activity
Dysfunction in neuronal network dynamics has been observed in the brain of patients with psychiatric and neurological conditions 27 . In addition, neuronal network dysfunction has been identified in model systems for several ID/ASD syndromes 20,28 . Therefore, despite the normal properties of KS iNeurons on single cell level, we hypothesized that impairments during brain development caused by loss of EHMT1 would be reflected on a higher level, i.e. at the network level. To test this hypothesis we examined and compared the spontaneous electrophysiological population activity of neuronal networks derived from controls and KS patients growing on microelectrode arrays (MEAs) ( Fig. 2A). MEAs allow us to non-invasively and repeatedly monitor neuronal network activity through extracellular electrodes located at spatially-separated points across the cultures.
First, we monitored the in vitro development of control neuronal networks by recording the MEAs in intervals of 1 week. We found that for control iNeuron networks the pattern of activity changed progressively over a period of several weeks ( Fig. 2B-F), similar to what we have observed previously in rodent neuronal cultures 29 . In particular, during the first three weeks of differentiation, controls, which was indicated by the smaller percentage of spikes occurring outside the network bursts (P=0.0077) (Fig. 2O). A final aspect is that, in comparison, the KS networks also exhibited an irregular network-bursting pattern, illustrated by the statistically larger coefficient of variation (CV) of the inter-burst interval (P<0.0001) (Fig. 2P). Interestingly, the increased burst duration phenotype observed at the network level was also present at the single-cell level ( Supplementary   Fig. S3P). Indeed, whole-cell voltage-clamp recordings of spontaneous excitatory postsynaptic currents (sEPSCs), show that the activity received by KS-derived neurons was composed by bursts with durations longer than in control. Taken together, our data show that KS neuronal networks consist of fewer and irregular network bursts, and the bursts themselves were longer in duration when compared to control networks.

CRISPR/Cas9 deletion of EHMT1 recapitulate KS patient phenotypes
To further address whether heterozygous loss of EHMT1 is necessary and/or sufficient for the observed KS patient-derived network phenotypes, we expanded our analysis and generated a second set of isogenic human iPS cells. We made use of CRISPR/Cas9 gene editing technology to generate an isogenic control and EHMT1 mutant iPS cell line with a premature stop codon in exon 2 (C CRISPR and KS CRISPR , Fig. 3A and Supplementary Fig. S5F, G). Western blot and RT-qPCR analysis revealed that EHMT1 expression was significantly (by more that 50%) reduced in KS CRISPR iPS (P=0.0079 for western blot; P=0.0054 for q-PCR) and iNeurons (P=0.0018) compared to C CRISPR In contrast to this, the KS CRISPR networks exhibited a phenotype similar to the other KS patient networks, where network bursts occurred at a lower frequency (P<0.0001), longer duration (P=0.0159) and in an irregular pattern (P=0.0346) compared to its isogenic control. These results, from two different isogenic iPS cell lines, including the iPS cells derived from the parent with a mosaic mutation (C MOS and KS MOS ), establish a causal role for EHMT1 in the observed neuronal network phenotypes.
Our in vitro experiments showed that EHMT1 deficiency causes a reproducible and identifiable neuronal network phenotype. As we observed only non-significant iPS cell line-specific variability in the functional network properties of, both, the control and KS groups, the multiple descriptive parameters we extracted from the raw MEA recordings clearly delineated control from KS networks. Indeed, we confirmed this in an unbiased discriminant analysis of network parameters, control and KS networks clearly clustered away from each other. This separation was not observed when the analysis was performed on single-cell parameters (i.e. morphology and intrinsic properties, Supplementary Fig. S6A-F). From a mechanistic perspective, our direct comparison of iNeurons derived from iPS cells with a frameshift, missense or deletion in EHMT1 showed that the phenotype is due to aberrant EHMT1 enzymatic activity rather than the disrupted protein. The activity and phenotype in these networks ultimately depend on many factors, such as ion channel and receptor expression, the number of functional synapses, and membrane electrical properties.
We next sought to uncover the contributing molecular pathways downstream of EHMT1, underlying the KS network phenotype.

Increased sensitivity to NMDA receptor antagonists in KS patient-derived neuronal networks
We observed that KS patient-derived neuronal networks showed an aberrant pattern of activity, mainly characterized by network bursts with longer durations compared to controls. Previous studies on rodent-derived neuronal networks have shown that burst duration is directly influenced by AMPARs and NMDA receptors (NMDARs). Specifically, previous reports used receptor-type specific antagonists to show that AMPAR-driven bursts have short durations while NMDAR-driven bursts have comparatively longer durations 30 . We therefore hypothesized that increased NMDAR activity contributed to the lengthened bursts seen in KS networks. To test this, we pharmacologically blocked either AMPARs or NMDARs and compared the effect on control and KS neuronal network activity at DIV 28. In accordance with previous work 30,31 , we found in our control networks that acute treatment with an AMPAR antagonist (NBQX, 50 µM) abolished all network burst activity, whereas inhibiting NMDARs (D-AP5, 60 µM) only slightly decreased the network burst activity (Fig. 4A, C). These results indicated that, in control neurons, network burst activity is mainly mediated by AMPARs. In particular, we found it to be mediated by GluA2containing AMPARs, since the network bursts were not blocked with Naspm (10 µM), an antagonist that selectively blocks GluA2-lacking AMPARs ( Supplementary Fig. S7B, pre-D-AP5).
Similar to controls, we found that in KS networks NBQX completely abolished network burst activity (Fig. 4B, D). However, in stark contrast to controls, D-AP5 robustly suppressed the bursting activity in all of our KS patient lines (Fig. 4B, D and Supplementary Fig. S7A), indicating that it was a common feature and a direct consequence of EHMT1-deficiency. Interestingly, the suppression of network burst activity by D-AP5 in KS networks was transient. Network burst activity showed ~50% recovery after 30 min and had returned to baseline (i.e. pre-D-AP5 levels) after 24 hours ( Fig. 4E and Supplementary Fig. S7C). The early stages of homeostatic plasticity, specifically synaptic upscaling, are initiated by global neuronal inactivity and characterized by insertion of GluA2-lacking (i.e. Ca 2+ -permeable AMPARs, CP-AMPARs) into the synapse to restore activity 32 . Therefore, to determine the nature of the recovery, we first blocked KS networks with D-AP5, and added Naspm after 1 hour. After 1 hour, Naspm completely blocked the reinstated network bursting activity, indicating the recovery in KS networks was due to synaptic insertion of GluA2-lacking AMPARs ( Fig. 4E and Supplementary Fig. S7B). Of note, although in control networks burst activity was not suppressed by D-AP5, we did observe that Naspm had a small but significant effect on the burst rate. This indicates that NMDAR blockage in controls also induced synaptic insertion of GluA2-lacking AMPARs, albeit less pronounced than in KS networks. After 24 hours, the network activity in KS networks was again completely suppressed with NBQX but only partially with Naspm, suggesting that GluA2-lacking AMPARs were actively exchanged with GluA2-containing AMPARs ( Supplementary Fig. S7C). Collectively, our results demonstrate that NMDAR inhibition with D-AP5 induces synaptic plasticity in KS networks, allowing reinstating network burst activity through the incorporation of GluA2-lacking AMPARs, which later on are replaced by GluA2-containing AMPARs.
Intrigued by these findings, we sought an alternative way to block network bursting independent of directly antagonizing AMPARs and NMDARs. The classic method for inducing synaptic upscaling, by preventing action potentials with the sodium channel blocker TTX 33 , would necessarily prevent observation of the early stages of the dynamic recovery on MEAs. To circumvent this issue, we used the anti-epileptic drug Retigabine, which is a voltage-gated K + -channel (K v 7) activator that effectively hyperpolarizes the resting membrane potential in neurons 34 . We reasoned that Retigabine would have the combined effect of acutely hampering neurons in reaching action potential threshold while leaving the voltage-gated Na + channels unaffected. Thus, while simultaneously strengthening the Mg 2+ block on NMDARs we could still observe any (re)occurring network activity on the MEA. Indeed, when we applied Retigabine (10 µM) to the networks, they were temporarily silenced, with no discernible spiking or bursting activity. Within 100 min, there was again a Naspm-sensitive recovery mediated by GluA2-lacking AMPARs, but this time it occurred in both the control and KS networks and the pattern in both was identical ( Supplementary   Fig. S7D). This reinforced the notion that the electrophysiological differences we saw earlier between control and KS networks are a direct consequence of NMDAR activity in the KS iNeurons.
Furthermore, we found that Retigabine treatment in KS networks decreased the burst length (P=0.012) ( Supplementary Fig. S7E), suggesting that Retigabine-induced plasticity allowed KS networks to switch from an NMDAR-dependent bursting to AMPAR-driven bursting, similar to controls.

NMDARs are upregulated in KS patient-derived neuronal networks
The results from our pharmacological experiments suggested that NMDAR expression might be increased in KS iNeurons relative to controls. To test this hypothesis, we measured the transcripts of the most common NMDAR and AMPAR subunits by RT-qPCR on total mRNA isolated from C MOS and KS MOS iNeurons (Fig. 5A). We were intrigued to find a 4-fold upregulation of GRIN1 mRNA, which encodes NMDAR subunit 1 (NR1), the mandatory subunit present in functional NMDARs (P<0.0001). We found no significant changes in any other NMDAR (GRIN2A and GRIN2B) or AMPAR (GRIA1, GRIA2, GRIA3 and GRIA4) subunits that we analyzed. We further corroborated these results with Western blot analysis which revealed significantly increased NR1 expression in lysates from KS MOS iNeurons compared to C MOS iNeurons (P=0.074) (Fig. 5B). Our previous functional data indicated that the reduction in methyltransferase activity of EHMT1 was directly responsible for the phenotypes we observed at the network level. Therefore, we investigated whether the increased GRIN1 expression correlated with reduced H3K9me2, the repressive mark mediated by EHMT1, at the GRIN1 promoter. To examine the levels of H3K9me2, we performed chromatin immunoprecipitation qPCR (ChIP-qPCR). We found that in KS MOS iNeurons H3K9me2 occupancy was reduced at the GRIN1 promoter (P=0.0148), but not at the promoter of the housekeeping gene PPIA (P=0.5337) (Fig. 5C). In accordance with our previous study in Ehmt1 +/mice 16 , we also found that the occupancy at the BDNF promoter was reduced in KS MOS iNeurons (P=0.0314) (Fig. 5C). Taken together, our data show that NR1 is upregulated in KS neuronal networks due to an aberrant regulation. Specifically, the loss of EHMT1 activity results in a reduction of the repressive H3K9me2 mark, resulting in an upregulation of NR1 and explaining the neuronal network phenotype we observed on the MEAs.

Altered neuronal network activity and increased NMDAR/AMPAR ratio in Ehmt1 +/mice
Having established that EHMT1-deficiency alters neuronal network activity due to an aberrant upregulation of NR1 in KS iNeurons, we next set out to measure neuronal network activity in Ehmt1 +/mice, a validated mouse model that recapitulates the core features of KS 14,15 . Similar to what we found in iNeurons, primary cultures of cortical neuronal networks derived from Ehmt1 +/mice showed network bursts with lower frequency (P=0.002) and longer duration (P=0.045) compared to cultures from littermate controls, whereas the mean firing rate was unaltered (P=0.297) ( Fig. 6A). Using whole-cell voltage clamp recordings in acute brain slices from WT and Ehmt1 +/mice, we measured the ratio between AMPAR-and NMDAR-mediated currents from cortical Layer 4 to Layers 2/3 synapses. We found that the NMDAR/AMPAR ratio was significantly increased in cortical networks of Ehmt1 +/mice as compared to WT littermates (P=0.02) (Fig. 6B). We found no changes in the kinetics of NMDAR-mediated currents, suggesting that there is no difference in the expression of NMDAR subunits 2A or 2B between cortical networks of WT and Ehmt1 +/mice ( Supplementary Fig. 6B) 35 . Because an increased NMDAR/AMPAR ratio can be caused by an upregulation of NMDARs, downregulation of AMPARs, or both, we measured AMPAR-mediated mEPSCs in Layers 2/3. We found no changes in the frequency (P=0.8441) or amplitude (P=0.5774) of AMPAR-mediated mEPSCs suggesting that the increased NMDAR/AMPAR ratio in Ehmt1 +/mice is due to increased NMDAR activity (Fig. 6C).

NMDAR inhibition rescues KS patient-derived neuronal network phenotypes
Our previous experiments showed that by inhibiting NMDARs in KS networks for 24 hours, we could shift the balance so that neuronal networks were progressively driven by GluA2-containing AMPARs, similar to controls ( Supplementary Fig. S7C). Based on these results, we reasoned that the phenotypes in KS networks could be rescued by chronically inhibiting NMDARs in mature neuronal networks (i.e. DIV 28). D-AP5, however, is a competitive antagonist that occupies the glutamate-binding site on NMDARs. Instead, we chose to potently block the channel pore of  (Fig. 7H). These results indicate that the aberrant pattern of activity exhibited by KS patient-derived neuronal networks can be rescued by specifically inhibiting NMDAR activity.

Discussion
In this study, we developed a human model of KS that enabled us to identify specific functional aberrations, from gene expression to neuronal network behavior, due to EHMT1 haploinsufficiency.
We found that excitatory networks derived from KS patients showed a distinct and robust neuronal network phenotype with striking similarity across different types of mutations in EHMT1. The phenotype was characterized by network bursts with a longer duration, lower frequency and more irregular pattern compared to controls. At the cellular level, we demonstrated that the network phenotype was mediated by upregulation of NR1.
Interestingly, the neuronal phenotype was consistent across species and model systems. Indeed, we found that network bursts also occurred with a lower frequency, longer duration and irregular pattern in dissociated neuronal networks from either embryonic rats (i.e. where Ehmt1 was downregulated through RNA interference 29 ). This indicates that some network parameters are consistently and similarly altered in divergent KS models. The appearance of a consistent phenotype both in a system where excitatory and inhibitory transmission were present and in a system where inhibition was absent (human iNeurons) indicates a major contribution from aberrant excitatory neurotransmission to KS pathobiology.
One of the major characteristics of KS networks was the longer duration of the network bursts compared to controls. A change in burst length can be indicative of synaptic changes in GABARs, NMDARs and/or AMPARs 31 . Given that inhibition was absent in our human model, we focused our analysis on NMDARs and AMPARs. We present several lines of evidence that suggest the long bursts exhibited by KS networks are mediated by the upregulation of NR1, the mandatory subunit for functional NMDARs. First, by acutely blocking AMPA and NMDA receptors with chemical inhibitors, we show that network burst activity in KS networks is strongly dependent on NMDARmediated transmission, in contrast to control networks, where network bursts are mainly dependent on AMPAR-mediated transmission. Second, we were able to reverse KS network phenotypes, including the long network burst duration, by blocking NMDAR activity in mature networks. Third, we found that NR1 is upregulated in KS iNeurons at both the mRNA and protein level. Fourth, NR1 upregulation is paralleled by H3K9me2 hypomethylation, an epigenetic modification caused by EHMT1 deficiency, at the GRIN1 promoter. Finally, we found increased NMDAR-mediated currents in the cortex of Ehmt1 +/mice, a mouse model for KS. This cross-species comparison further validates that the effects we observed are due to decreased EHMT1 enzymatic function and show that the development of network activity in human and mouse cortical networks may follow evolutionarily conserved and stable epigenetic programming.
Genetic evidence has directly implicated NMDARs directly in NDDs. For example, multiple heterozygous mutations in NMDAR subunit genes have been identified to be as causal for ID, ASD or epilepsy 37 . More indirectly, NMDAR dysfunction has been observed in humans and rodent models of ID/ASD. Although in most cases NMDAR dysfunction is attributed to hypofunction, there are several observations associating NMDAR hyperfunction to ID/ASD 38 . For example, upregulated NR1 protein levels were found in the cerebellum of ASD patients 39 . Furthermore, the NR2A and NR2B subunits of the NMDA receptor are upregulated in the valproic acid (VPA) animal model of autism 40 . In the Rett syndrome mouse model 41 , loss of MECP2 function resulted in developmental dysregulation of NMDAR expression 35,42 . Of note, work in mouse models identified temporal and spatial changes in NMDARs, illustrating the importance of age and brain regions in evaluating NMDAR expression in different genotypes 43,44 . Our data showing that NR1 is upregulated in KS iNeurons of cortical identity is the first to present NMDAR hyperfunction in excitatory neurons derived from patients with ID/ASD and supports the notion that glutamatergic neurotransmission can be dysfunctional in some NDDs.
By blocking NMDARs in KS networks, we were able to induce synaptic plasticity (i.e the early phase of synaptic upscaling) that enabled, first, the incorporation of GluA2-lacking receptors, followed by the insertion of GluA2-containing receptors. Interestingly, this plasticity mediated by AMPARs was more easily initiated in KS than in control networks and allowed KS networks to switch, at least temporarily, from a mainly NMDAR-driven network to an AMPAR-driven network.
These results suggest that loss of EHMT1 could facilitate NMDAR-mediated plasticity after a comparatively milder stimulus, in this case NMDAR blockade with D-AP5. This is in agreement with a recent study showing that inhibition of EHMT1/EHMT2 activity reinforces early-LTP in an NMDAR-dependent manner 18 . The authors showed that pharmacologically blocking EHMT1/2 before a mild LTP stimulus increased the LTP response, highlighting a role for EHMT1 and associated H3K9me2 in metaplasticity 18 . A link between NMDARs and EHMT1 has also been shown in vivo where NMDAR activity regulates the recruitment of EHMT1/2 and subsequent H3K9me2 levels at target gene promoters in the lateral amygdala, in the context of fear learning 45 .
This, together with our data, suggests that there is a reciprocal interaction, between NMDAR activity and EHMT1 function, and a positive feedback mechanism where EHMT1 methylates the NR1 promoter upon activation by NMDARs. If and under which circumstances such a feedback mechanism is active during development and/or in adulthood deserves further investigation.
At the molecular level we show that the upregulation of NR1 correlates with reduced H3K9me2 occupancy at the GRIN1 promoter. This suggests that during normal development EHMT1 is directly involved in the regulation of NR1 expression through H3K9me2 deposition at the GRIN1 promoter. However, we cannot exclude that EHMT1 also regulates NR1 expression via other, less direct mechanisms. For example, EHMT1 is a member of a large complex that includes the Neuron-Restrictive Silencer Factor (NRSF/REST) repressive unit, which is important for repressing neuronal genes in progenitors, including NR1 [46][47][48] . In addition, growth factors such as BDNF, which is increased upon EHMT1 deletion 16 , have been shown to increase NR1 mRNA levels in cultured embryonic cortical neurons 49 .
An important aspect of our study is the identification of a robust and consistent network phenotype linked to KS and measured with MEAs. We show that KS networks differed from controls based on a set of parameters describing the neuronal network activity and using discriminant analysis. Our analysis showed that individual unrelated controls clustered with little variation, which was exemplified by the fact our predictive group membership analysis was able to accurately assign control networks to the control group (Fig. S6). In contrast, we found that KS patient lines significantly differed from controls, including their respective isogenic controls (C MOS vs K MOS and C CRISPR vs KS CRISPR ). In addition, when we performed predictive group membership analysis, we found that individual KS networks were mostly assigned to the corresponding patient line (Fig. S6), indicating that this level of analysis on neuronal networks can potentially detect patient-specific phenotypic variance that arises early in development. It is foreseeable that, in future studies, an in depth interrogation of the network activity networks, consisting of different human derived cell types (inhibitory, astrocytes, microglia,…) or networks from brain organoids would allow measuring more complex neuronal signals on MEAs. This would be especially relevant for the stratification of genetically complex disorders (e.g., idiopathic forms of ASD) as these in vitro network phenotypes could then be used as an endophenotype in pharmacological studies.
We show it is possible to rescue the neuronal phenotype by blocking NMDA receptors in mature networks, a finding that has important clinical relevance. For example, NMDAR antagonists, such as ketamine and memantine, have been used successfully as treatment strategies in mouse models of RTT and have led to improvements in small open-label trials for autism [50][51][52] . These studies, as well as ours, provide preclinical proof of concept that NMDAR antagonists could ameliorate neurological dysfunction and reverse at least some circuit-level defects. Furthermore, our observation that neuronal phenotypes can be rescued in mature networks agrees with previous data showing that reinstating EHMT1 function in adult flies is sufficient to rescue memory deficits 13 .
This adds to a growing list of genetically defined ID syndromes that might be amenable to postnatal therapeutic intervention 53 .
In summary, our study shows that combining iPS cell-derived human neuronal models with neuronal network dynamics is a promising tool to identify novel targets that inform us about possible treatment strategies for NDDs such as ID and ASD.

Patient information and iPS cell generation
In this study we used in total four control and four iPS cells with reduced EHMT1 function. In contrast to a previous study 54 we included patients in this study that present the full spectrum of KS associated symptoms, including ID and ASD. KS 1  Medium was refreshed every 2-3 days and cells were passaged twice per week using an enzymefree reagent (ReLeSR, Stem Cell Technologies).

CRISPR/Cas9 editing of EHMT1
We made use of the CRISPR/Cas9 technology in order to create a heterozygous EHMT1 mutation in Exon 2 in a iPS cell line derived from a healthy 51 year-old male to mimic KS in isogenic cell lines, generated by KULSTEM (Leuven, Belgium). In brief, 1x10 6 Fig. S5). The generated iPS cells were validated for pluripotency markers and quantitative analysis of tri-lineage differentiation potential was performed. All generated iPS cell lines have the capacity to differentiate towards all three germ layers (endoderm, mesoderm, ectoderm). To this end embryoid bodies were generated in 24-well Corning low attachment plates.
For spontaneous differentiation, the culture was kept for 7 days in E6 medium (Thermo Fisher Scientific). The medium was changed every 2 days. Cells were harvested after 7 days for RNA extraction with the GenElute Mammalian Total RNA kit (Sigma). cDNA synthesis was performed with Superscript III and used for qPCR according to manufacturer's protocol with TaqMan human iPS cell Scorecard assay (Life Technologies). Data analysis was performed with Scorecard software (online tool Life technologies), comparing with a reference set of pluripotent stem cell lines.

Western Blot
For Western Blot cell lysates were made from confluent wells in 6-well plates of either iPS cells or iNeurons. Medium was always refreshed 4 hours beforehand. Protein samples were loaded, separated by SDS-PAGE, and transferred to nitrocellulose membrane (BIO-RAD). The membrane was then probed with an EHMT1 antibody (Abcam, ab41969) or NMDAR1 (Biolegend, 818601).

Neuronal differentiation
iPS cells were directly derived into, excitatory cortical Layer 2/3 neurons by overexpressing the neuronal determinant Neurogenin 2 (NGN2) upon doxycycline treatment based on Zhang et al. 24 and as we described previously 25 . To support neuronal maturation, freshly prepared rat astrocytes 25 were added to the culture in a 1:1 ratio two days after plating. At days in vitro (DIV) 3 the medium was changed to Neurobasal medium (Thermo Fisher Scientific) supplemented with B-27 (Thermo Fisher Scientific), glutaMAX (Thermo Fisher Scientific), primocin (0.1 µg/ml), NT3 (10 ng/ml), BDNF (10 ng/ml), and doxycycline (4µg/ml). Cytosine b-D-arabinofuranoside (Ara-C) (2µM; Sigma) was added once to remove any proliferating cell from the culture. From DIV 6 onwards half of the medium was refreshed three times a week. The medium was additionally supplemented with 2,5% FBS (Sigma) to support astrocyte viability from DIV 10 onwards. Neuronal cultures were kept through the whole differentiation process at 37°C/5%CO 2 .

Cortical cultures from mice
Primary cortical neurons were prepared from Ehmt1 +/and Ehmt1 +/+ mice from individual E16.5 embryos as previously described 16 . Since the genotype was unknown at the time of harvest, each embryo was collected and the brains were processed separately. Each whole brain was kept on ice in 1 mL L-15 medium, organized separately in a 24-well plate, and tail clips were collected for genotyping.

Neuronal morphology reconstruction
To examine morphology of neurons cells on coverslips were transfected with plasmids expressing Discosoma species red (dsRED) fluorescent protein one week after plating. DNAin (MTI-GlobalStem) was used according to manual instructions. Medium was refreshed completely the day after DNAin application. After the treatment cells were cultured as previously described.

Micro-electrode array recordings and data analysis
All recordings were performed using the 24-well MEA system (Multichannel Systems, MCS GmbH, Reutlingen, Germany). MEA devices are composed by 24 independent wells with embedded micro-electrodes (i.e. 12 electrodes/well, 80 µm in diameter and spaced 300 µm apart).
Spontaneous electrophysiological activity of iPS cell-derived neuronal network grown on MEAs was recorded for 20 min. During the recording, the temperature was maintained constant at 37°C, and the evaporation and pH changes of the medium was prevented by inflating a constant, slow flow of humidified gas (5% CO 2 and 95% O 2 ) onto the MEA plate (with lid on). The signal was sampled at 10 KHz, filtered with a high-pass filter (i.e. Butterworth, 100 Hz cutoff frequency) and the noise threshold was set at ± 4.5 standard deviations.
Data analysis was performed off-line by using Multiwell Analyzer (i.e. software from the 24-well MEA system that allows the extraction of the spike trains) and a custom software

Chemicals
All reagents were prepared fresh into concentrated stocks as indicated below, and stored frozen at

Fig. 2. Spontaneous electrophysiological activity of control-and KS patient-derived neuronal networks. A)
Representative image of a control-derived neuronal network on MEAs stained for MAP2 (red) and PAN Axon (green).   E) The effect of a 1-hour D-AP5 treatment on C MOS and KS MOS neuronal network activity. After 1 hour, the calciumpermeable AMPA receptor blocker Naspm (10 µM) or NBQX were added. Data represent means ± SEM. ** P<0.005, *** P<0.0005, n=8 for C MOS and n=9 for KS MOS , one-way ANOVA test followed by a post hoc Bonferroni correction was performed between conditions.