Loss of CNTNAP2 Alters Human Cortical Excitatory Neuron Differentiation and Neural Network Development

BACKGROUND: Loss-of-function mutations in the contactin-associated protein-like 2 ( CNTNAP2 ) gene are causal for neurodevelopmental disorders, including autism, schizophrenia, epilepsy, and intellectual disability. CNTNAP2 encodes CASPR2, a single-pass transmembrane protein that belongs to the neurexin family of cell adhesion molecules. These proteins have a variety of functions in developing neurons, including connecting presynaptic and postsynaptic neurons, and mediating signaling across the synapse. METHODS: To study the effect of loss of CNTNAP2 function on human cerebral cortex development, and how this contributes to the pathogenesis of neurodevelopmental disorders, we generated human induced pluripotent stem cells from one neurotypical control donor null for full-length CNTNAP2 , modeling cortical development from neurogenesis through to neural network formation in vitro. RESULTS: CNTNAP2 is particularly highly expressed in the ﬁ rst two populations of early-born excitatory cortical neurons, and loss of CNTNAP2 shifted the relative proportions of these two neuronal types. Live imaging of excitatory neuronal growth showed that loss of CNTNAP2 reduced neurite branching and overall neuronal complexity. At the network level, developing cortical excitatory networks null for CNTNAP2 had complex changes in activity compared with isogenic controls: an initial period of relatively reduced activity compared with isogenic controls, followed by a lengthy period of hyperexcitability, and then a further switch to reduced activity. CONCLUSIONS: Complete loss of CNTNAP2 contributes to the pathogenesis of neurodevelopmental disorders through complex changes in several aspects of human cerebral cortex excitatory neuron development that culminate in aberrant neural network formation and function.

CNTNAP2 encodes a single-pass transmembrane protein, also known as CASPR2, which belongs to the neurexin family of cell adhesion molecules (40).While the precise molecular functions of CNTNAP2 remain poorly understood, neurexins serve a variety of functions in the developing central nervous system, including mediating synaptic signaling (41).Additionally, CNTNAP2 interacts with contactin-2 (CNTN2), forming a complex thought to be required for the clustering of voltagegated potassium channels (42).CNTNAP2 LOF mutations are proposed to contribute to neurodevelopmental disorders through alterations in synaptic transmission, neuronal connectivity, and/or network-level activity.Consistent with this, complete loss of Cntnap2 in the developing mouse reduces neurite branching and dendritic spine density in cortical neurons (43)(44)(45)(46)(47)(48).
To study how loss of CNTNAP2 function contributes to the pathogenesis of neurodevelopmental disorders in the developing human forebrain, we used CRISPR (clustered regularly interspaced short palindromic repeats)-Cas9 to generate human CNTNAP2-null induced pluripotent stem cells (iPSCs) from one neurotypical control donor.We used this system to replay cortical development from neurogenesis to neural network formation (49,50).We find that loss of CNTNAP2 significantly alters human cortical excitatory neuron development, indicating that the pathogenesis of neurodevelopmental disorders due to CNTNAP2 LOF begins early in development.

METHODS AND MATERIALS
Full experimental methods are given in Supplemental Methods.A summary of the number of cell lines and replicates used in each experiment is reported in Table S7.

Human PSC Culture and Neural Differentiation
PSCs were differentiated to cerebral cortex progenitor cells, according to our reported methods (49).

Single-Cell Messenger RNA Sequencing
CNTNAP2 null and isogenic control PSC-derived cortical differentiations, 50 days after neural induction (D50), were dissociated using the Papain Dissociation System and multiplexed using cell hashing antibodies, as previously described (53).Single-cell sequencing was performed using the 10x Genomics Single Cell 3 0 kit version 2, and count matrices were analyzed using the Seurat (version 4; Paul Hoffman, Satija Laboratory and Collaborators) package.Differential expression between null and control samples was performed using DESeq2 (Bioconductor) (54) on pseudobulk-aggregated counts.

Neurite Length and Branching Measurement
CNTNAP2 null and isogenic control neurons were sparsely nucleofected with a fluorescent construct at D50 using an Amaxa 4D (Rensselaer Polytechnic Institute) nucleofector.Cultures were subsequently imaged on an Opera Phenix confocal microscope every 3 to 6 days until D61 or D66.Following image acquisition, neurite length and branching were measured with NeuronStudio (Lumitos AG) software (55).

Calcium Imaging and Multielectrode Array Analyses
Calcium imaging was conducted using the Incucyte S3 (Thermo Fisher Scientific).For multielectrode array (MEA) analyses, activity was recorded on a Maestro Pro MEA system, using the accompanying Axis Navigator software.

CNTNAP2 Is Highly Expressed in Differentiating Cortical Excitatory Neurons In Vitro and In Vivo
To determine the developmental expression of CNTNAP2, we measured CNTNAP2 messenger RNA (mRNA) and protein throughout cortical differentiation from iPSCs.For this, we used our established method to differentiate human iPSCs to cerebral cortex neural progenitor cells and glutamatergic excitatory cortical neurons (Figure 1A) (49,50,56).With this method, neurons are generated in the same temporal order and over a comparable timeframe as observed in utero (49,50,56).iPSC-derived neurons also display similar gene expression profiles to their primary counterparts (Figure 1B; Figure S1) (49,50,57).
There are 4 predicted protein-coding isoforms of CNTNAP2 (58-60), including the canonical full-length transcript referred to as CNTNAP2-201 (Figure 1C).During cerebral cortex development there is clear evidence for the expression of CNTNAP2-201 as well as a short isoform, CNTNAP2-203, which is located at the 3 0 end of the locus and encodes a predicted 12-kDa protein (Figure S2).Because the majority of known disease-associated mutations affect only the full-length transcript, we focused the functional studies reported here on that CNTNAP2 isoform (38,61).
CNTNAP2 expression was measured by quantitative reverse transcriptase polymerase chain reaction at 4 stages of the cortical differentiation process: 1) PSCs (i.e., predifferentiation), 2) cortical neural progenitor cells (D30), 3) early-born deep layer neurons (D50), and 4) late-born deep layer neurons (D80).The latter time point also includes the period of maturation of deep layer neurons that would have been generated over the previous 30 days.Full-length CNTNAP2 mRNA levels significantly increased during the cortical differentiation process, with no expression detected in iPSCs (one-way repeated-measures analysis of variance [ANOVA]: F 3,24 = 167.9,p = 3.1 3 10 216 ) (Figure 1D).CNTNAP2 mRNA was first detected at D30, when most cells are cortical progenitor cells and was significantly more highly expressed than iPSCs (p = 1.0 3 10 23 for iPSC vs. D30).Expression of CNTNAP2 significantly increased between D50 and D80 (p = 4.1 3 10 25 for D50 vs. D80).Therefore, CNTNAP2 is most highly expressed in postmitotic neurons, increasing over time, and is also expressed in neural progenitor cells but at a significantly lower level.
Comparison of the in vitro time course of CNTNAP2 expression with a publicly available RNA sequencing (RNA-Seq) dataset of primary fetal cortex (62) indicated that in vivo developmental expression follows equivalent temporal and cell type patterns.Both long and short CNTNAP2 isoforms are present as early as postconception week 8 (PCW 8) in human prefrontal cortex (Figure 1E) (earlier data are not currently available).Expression then rises until PCW 12, which would match the increased CNTNAP2 expression between D50 to D80 in vitro.Cortical neurogenesis begins in human embryos around PCW 7 (63,64), thus the onset of CNTNAP2 expression at PCW 8, and its subsequent increase, is consistent with most expression being in postmitotic neurons.
Expression of the long isoform of CNTNAP2 protein follows the same temporal progression during iPSC differentiation as its mRNA, first becoming detectable in cortical progenitor cells (D30) and increasing thereafter (ANOVA: F 3,9 = 55.8, p = 3.9 3 10 26 ; pairwise t tests: p = .008for iPSC vs. D50; p = .03for D30 vs. D80) (Figure 1F).Immunofluorescence and confocal microscopy of iPSC-derived neurons found that CNTNAP2 is enriched in discrete puncta along axons, dendrites, and neuronal cell bodies (Figure 1G).Puncta colocalized with both presynaptic (SYP 1 ) and postsynaptic markers (PSD-95 1 ), indicating that CNTNAP2 is present at the neuronal synapse, as previously reported in mice (48).To generate human iPSCs with loss-of-function (LOF) alleles in CNTNAP2, a guide RNA was designed to disrupt the gene at exon 3 in the NDC1.2 cell line (Figure S3A) (52).Successful targeting was confirmed by MiSeq next generation sequencing (Illumina) and Sanger sequencing of genomic DNA.This sequencing revealed insertion of the stop codon in both alleles of edited iPSC clones (Figure S3B, C).Comparison of cell fate patterning between null and control inductions was performed using the Nanostring nCounter Sprint Profiler, using a customdesigned human brain development code set of 150 genes  (56).Results from this analysis showed that null and control cultures were not significantly different in composition (Welch's t test corrected for multiple testing: p ..05) (Figure S3D; Table S2).Finally, specific LOF of the long isoform of CNTNAP2 was confirmed at the level of mRNA and protein in D50 cortical neurons by quantitative reverse transcriptase polymerase chain reaction and Western blot (Figure S3E, F).
To gain insights into both cell composition and cell typespecific gene expression changes as a consequence of CNTNAP2 LOF during human cortical development, single-cell RNA-Seq (scRNA-Seq) was carried out on 2 independent cortical differentiations from CNTNAP2 null and isogenic control iPSCs (Figure S4).Single-cell analyses were performed using the 10x Genomics platform at D50, a point of large-scale network formation (57) and high CNTNAP2 protein expression.A selection of cell type-associated marker genes (Figure S5) was used to identify clusters corresponding to 7 broad cell types, including cortical excitatory neurons and progenitors (Figure 2A, B).
We used the single-cell dataset to investigate cell typespecific expression of the long isoform of CNTNAP2 in cortical development (Figure S6).Because the scRNA-Seq approach used here only detects the 3 0 end of transcripts, which, in the case of CNTNAP2 is shared by the long and short isoforms, it cannot distinguish between the different transcripts (58)(59)(60)65).However, quantitative reverse transcriptase polymerase chain reaction showed that targeting the 5 0 exons of the gene specifically reduced full-length CNTNAP2 expression (Figure S3E).Therefore, we examined differential expression of CNTNAP2 between control and CNTNAP2 null cell types, finding that cortical neurons were the only cell type in which total CNTNAP2 expression was significantly reduced in null differentiations (Figure 2C).This suggests that the long form of CNTNAP2 is primarily expressed in neurons.Notably, cortical neurons also exhibited the highest number of differentially expressed genes (Figure 2D), indicating that the loss of CNTNAP2 directly leads to altered gene expression in those cells.
To investigate the effects of CNTNAP2 LOF on cortical differentiation at single-cell resolution, the subset of cells that was identified as cortical progenitors and neurons was extracted for further analysis (4917 cells out of the total 11,941 analyzed).To assign predicted differentiation paths in an unbiased manner, we first ordered cells using a pseudotime algorithm (66), which identified a single trajectory from progenitor cell to differentiating neurons (Figure 2E, F).
Comparing gene expression between CNTNAP2 null and isogenic control cortical neurons identified several mRNAs with large differences in gene expression (Figure 2G; Table S3), most notably, increased expression of calretinin (CALB2) and neurotensin (NTS) in CNTNAP2 null neurons, and decreased expression of LMO3, NEUROD2, and CSRP2.Inspecting the expression of these genes along the pseudotime trajectory, CALB2 and NTS are expressed in control neurons before LMO3, although this trend is less clear for NEUROD2 and CSRP2 (Figure 2H).In CNTNAP2 null neurons, CALB2 expression begins earlier than in controls, with LMO3 delayed relative to controls.

Loss of CNTNAP2 Results in Altered Differentiation of Cortical Neuron Subtypes
To further analyze the significance of the alterations in gene expression in postmitotic cortical neurons, the subset of cortical neurons in the dataset were extracted and reclustered (Figure 3A).Four clear populations were identified, representing differentiating neurons, early postmitotic neurons, and 2 well-separated populations of terminally differentiated neurons, referred to here as the late 1 and 2 groups (Figure S7; Table S4).
The 2 genes highly differentially expressed between CNTNAP2 null and isogenic control neurons, LMO3 and CALB2, mapped robustly onto the late 2 and late 1 clusters, respectively, as did NEUROD2 and NTS (Figure 3B).This indicates that the differential expression between CNTNAP2 null neurons and controls most likely reflects a change in the proportions of these 2 neuronal types.Consistent with this, we found a reduction in LMO3 1 , late 2 neurons in CNTNAP2 null compared with controls, and an accompanying increase in CALB2 1 , late 1 neurons (Figure 3C, D).This increase was not due to simple experimental variation during cortical differentiation, because it was found in 2 independent cortical inductions from iPSCs.
The 2 cell types that show significant reductions in CNTNAP2 expression in the null differentiations are the late 1 and late 2 populations, with the greatest reduction in expression observed in the LMO3 1 /late 2 population (Figure 3E).In addition to changes in cell type composition, we also analyzed differential gene expression between CNTNAP2 null and isogenic control CALB2 1 /late 1 and LMO3 1 /late 2 neurons (i.e., within each population) (Figure 3F, G; Table S5).For both populations, CALB2 and NTS were highly upregulated in the CNTNAP2 null, and LMO3 was downregulated.Additionally, gene set enrichment analysis identified there was a significant downregulation of a set of genes that all encode elements of the cholesterol and fatty acid synthesis pathways, including HMGCR and HMGCS1 (Table S6).In primary fetal cortex, these genes are most highly expressed in neurons of early and intermediate maturity, with expression reduced in late neurons (67).Thus, these data are consistent with an overproduction of early CALB2-expressing neurons and an underproduction of later differentiating LMO3 1 neurons.

Reduced Neurite Branching in Human CNTNAP2 Null Cortical Neurons
Previous studies in mice null for CNTNAP2 identified changes in neurite number and branching (43)(44)(45).Therefore, we investigated whether loss of CNTNAP2 affects neurite outgrowth from human cortical excitatory neurons.We utilized constructs at day 35, which birthdated the nucleofected neurons as a cohort.Randomly selected neurons were imaged live from wD40 until wD65, and neuron reconstruction software (55) was used to measure 3 parameters: neurite branching, neurite length, and total neuron length (Figure 4B).Example images of D60 neurons are provided in Figure 4C.
For mixed cultures, a significant reduction in the number of branches per neuron was detected in CNTNAP2 null cells Genes regulating cortical progenitor cell maintenance (SOX2 and PAX6) show the highest expression at the beginning of the trajectory, and genes known to increase with neuronal maturation (MAPT and DCX) show the highest expression toward the end.The expression patterns of these genes do not differ between CNTNAP2 null and isogenic control samples.Arrows under the x axes indicate the pseudotime values associated with cortical progenitors (purple) and cortical neurons (green).(G) Differentially expressed genes between CNTNAP2 null and control cortical neurons.The volcano plot shows the log 2 FC and associated p value for each gene, with genes passing statistical significance highlighted in red (adjusted p value , .05)(78 genes).Genes with a negative log 2 FC have higher expression in control neurons and those with a positive FC have higher expression in null neurons.(H) Differentially expressed genes between CNTNAP2 null and control cortical neurons show the highest expression toward the end of the cortical trajectory.Relative expression and generalized additive model-fitted lines for CALB2 and NTS (higher expression in null neurons) as well as LMO3, NEUROD2, and CSRP2 (higher expression in control neurons) is shown ordered by cortical pseudotime, as in (F).(Figure 4D) (two-way repeated-measures ANOVA; separate cultures: F 1,17 = 0.79, p = .39;mixed cultures: F 5,175 = 14.54, p = 6.51 3 10 212 ).While ANOVA for separate cultures did not reach significance, post hoc comparisons did identify CNTNAP2 null cells as having a significantly reduced number of branches at D55 (p = .025after Bonferroni correction).Significant reductions were also found at D55, D60, and D66 in the mixed cultures (by both ANOVA and post hoc analyses) (post hoc: p = .015,p = 5 3 10 26 , and p = .041,respectively).For average neurite branch length, no significant difference was detected between control and CNTNAP2 null neurons at any time point in the separate genotype experiment (Figure 4E) (two-way repeated-measures ANOVA: F 1,17 = 0.41, p = .53).This was confirmed by post hoc analyses, in which p ..05 for all comparisons.In the mixed cultures, repeated-measures ANOVA also did not reach significance (ANOVA F 1,35 = 1.1, p = .3).However, post hoc pairwise comparisons (corrected for multiple testing) did find CNTNAP2 null neurons to have significantly increased branch length at D48 and D60 (p = .017and p = .006).
Finally, CNTNAP2 knockout did not affect total neuron length in genotype-specific cultures (Figure 4F) (two-way repeated-measures ANOVA: F 1,17 = 0.079, p = .78,post hoc:  , late 1 neurons is increased in the CNTNAP2 null samples, while the density of LMO3 1 , late 2 neurons is reduced.In (D) the percentage of all cortical neurons belonging to each subpopulation is plotted (separated by genotype).Error bars represent standard error of the mean.(E) Full-length CNTNAP2 is expressed in the most mature neuronal populations (late 1 and 2).Differential expression of CNTNAP2 was used as a proxy for full-length expression as in (C).Log 2 FC between CNTNAP2 null and control samples in the indicated population is shown as a heatmap with comparisons not passing statistical significance (adjusted p value $ .05) in gray.(F, G) Differential gene expression between CNTNAP2 null and control samples in late 1 (F) and late 2 (G) cortical neurons.Volcano plots show the log 2 FC and p value for each gene, with genes passing statistical significance (adjusted p value , .05)highlighted in red.In both late 1 and late 2 cortical neurons, CALB2 and NTS were highly upregulated in the CNTNAP2 null samples, while LMO3 was downregulated.Additionally, downregulation of multiple genes belonging to the cholesterol and fatty acid synthesis pathways, including SQLE, FDFT1, DHCR7, HMGCR, and HMGCS1, was also observed in CNTNAP2 null neurons from both late 1 and late 2 subtypes.Expression of example cholesterol and fatty acid synthesis genes in late 1 and late 2 neurons is shown as a dot plot separated by sample of origin.Dot color represents scaled expression levels and dot size the percentage of cells with nonzero expression.Ctrl, control; log 2 FC, log 2 fold change; UMAP, Uniform Manifold Approximation and Projection.S8).Out of the 40 time points examined, post hoc comparisons found that the majority had a significant difference in activity between genotypes.Between wD41 and D60, control neurons had a greater number of active objects and a higher burst rate (Sidak's multiple comparisons test; active objects: p , .003;burst rate: p , .007).From D63 onward, there was a marked switch in activity and a larger magnitude difference emerged, with null neurons being significantly more active than controls, both in terms of more active objects (p , .0007) and a higher burst rate (p , .002).Notably, while activity increased in control neurons at a relatively stable rate over weeks in culture, activity in CNTNAP2 null neurons switched from increasing to decreasing from wD70 onwards, eventually reducing to control levels.
Recording activity with MEAs revealed similar trends (Figure 5D) (two-way ANOVA; number of spikes: F 16,451 = 1.57, p = .07;burst frequency: F 16,451 = 3.36, p , .0001).Control neurons initially had more spikes and a higher burst frequency per active electrode than CNTNAP2 null neurons until wD60.Due to the high degree of variability in the MEA readings, however, these comparisons were not significant by post hoc analysis.From D60 until D77, null neurons recorded more spikes and a higher burst frequency per active electrode (Welch's t test; number of spikes: D60: p = .004;D72: p = .03;and D77: p = .02;burst frequency: D72: p = .03;D77: p = .008).Again, activity steadily increased throughout the experiment in control neurons but began to decrease from wD80 in proteins localized to dendrites (MAP2, green), axons (tau, red), and cell bodies (nuclei labeled with DAPI 1 , blue).Scale bar = 100 mm.(C) CNTNAP2 null neurons and networks initially show lower activity than isogenic controls, as measured by calcium imaging.This is then followed by an extended (.30 days) period of significant hyperactivity in CNTNAP2 null neurons.Activity was recorded every 24 to 48 hours and the average number of active neurons and average burst rate were calculated.Active objects are defined as cells or cell clusters that fire at least once over a 2-minute scan.From wD60 to D90, the number of active objects and the average burst rate were significantly increased in CNTNAP2 null neurons followed by a decrease in activity to a level equal to or below

DISCUSSION
We report here that complete loss of function of the long isoform of CNTNAP2 has multiple effects on cortical development.Comparing iPSCs null for full-length CNTNAP2 with isogenic controls from the same donor cell line, we found that loss of CNTNAP2 leads to specific defects in the generation and differentiation of the first-born neurons in the cortex.Furthermore, loss of CNTNAP2 leads to cell autonomous defects in neurite outgrowth and branching, reducing excitatory neuron dendritic arbor complexity.Loss of CNTNAP2 significantly alters cortical excitatory network function, manifesting in an initial period of reduced activity, a subsequent lengthy phase of developing network hyperexcitability, followed by a reduction in spontaneous activity below control network levels.
To study the consequences of loss of full-length CNTNAP2 for human cortical development, we used scRNA-Seq to compare both cell composition and gene expression between CNTNAP2 null and isogenic control human iPSC-derived cortical progenitors and neurons.Within cortical neurons, the most significantly differentially expressed genes included downregulation of LMO3, CSRP2, and NEUROD2 in null neurons and upregulation of CALB2 and NTS.This likely reflects changes in neuronal subpopulations-both in terms of gene expression and relative proportions.CNTNAP2 null differentiations had a smaller proportion of cells in the LMO3 1 /late 2 population (which highly expresses the downregulated genes) and a higher proportion in the CALB2 1 /late 1 population (which highly expresses the upregulated genes).Unfortunately, with 2 independent scRNA-Seq experiments, there is not sufficient power to assign statistical significance to changes in cell proportions, whereas the experiments are powered to detect significant changes in gene expression.
While we were not able to conclusively say whether the LMO3 1 /late 2 and CALB2 1 /late 1 neurons are two distinct terminally differentiated neuronal populations, or whether one population is a transition stage to another, there is evidence suggesting that the CALB2 1 /late 1 population may be precursors of the LMO3 1 /late 2 group.By Carnegie stage 21, the human pioneer cortical plate contains primarily CALB2 1 / TBR1 1 neurons (68).CALB2 1 pioneer neurons disappear shortly after the formation of the definitive cortical plate, which is established around Carnegie stage 21/22 when TBR1 1 / CALB2-negative neurons appear.
In contrast, LMO3 expression is almost entirely restricted in its onset of expression to the cortical subplate, which forms at a later developmental stage than the pioneer cortical plate (69).One possibility is that the CALB2 1 /late 1 population are presubplate cells that subsequently downregulate CALB2 and upregulate LMO3.This would suggest that the CNTNAP2 null cells are delayed or retained in the CALB2-high stage, rather than progressing on to an LMO3-high stage.Alternatively, loss of CNTNAP2 could lead to an increased production of CALB2 1 early neurons from cortical progenitor cells.
Additionally, we analyzed neurite development in both CNTNAP2 null and isogenic control excitatory neurons.CNTNAP2 null neurons exhibit defects in neurite outgrowth, with the number of branches per neuron reduced compared with controls.Analysis of CNTNAP2 null neurons cocultured with isogenic control neurons demonstrated that the reduction in neurite branching is a cell autonomous phenotype.The reduction in branch number was also more significant in the cocultures.We believe that this is due to the coculture technique controlling for variability between extracellular environments.Such variability may have obscured some of the reduction in branching and/or neurite length in the singlegenotype cultures.
Finally, in addition to neuronal morphogenesis, we also studied the effect of CNTNAP2 LOF on excitatory neural network development, using calcium imaging and MEAs.Both approaches demonstrated that CNTNAP2 null cortical networks were significantly more active than those of isogenic controls, beginning around D60, before reverting to control levels and eventually demonstrating reduced activity.The early hyperactivity phenotype is consistent with previous reports of human forebrain networks generated from iPSCs with a heterozygous CNTNAP2 mutation (derived from an individual with schizophrenia) (70).
Recent work has suggested that CNTNAP2 acts to promote calcium export from neurons, thereby suppressing excitability and network activity (71).A loss of CNTNAP2 could therefore lead to hyperexcitability by increasing intracellular calcium, a hypothesis that should be tested in future experiments.Such a mechanism could also explain the reduction in neurite branching that we observed.Neurite outgrowth may be suppressed to compensate for the heightened activity of the null cells.
Altered neuronal activity has been reported in postnatal Cntnap2 null mice, in which Cntnap2 null neurons were hypoactive (47,48,(72)(73)(74).Midgestation and prenatal cortical network activity in these models have not been reported; however, the reduced postnatal activity in mouse models is consistent with the late-stage phenotypes of reduced human cortical network activity reported here.Furthermore, while our model system does not produce cortical interneurons, there is robust evidence that alterations in GABAergic (gamma-aminobutyric acidergic) signaling also occur in the context of reduced CNTNAP2 dosage (75,76).Further investigation into cortical interneuron dynamics in a human CNTNAP2 null model should also be conducted.
We conclude that CNTNAP2 LOF mutations change multiple aspects of cortical development, ranging from neuronal subtype differentiation to the formation of neural circuitry.Because these phenotypes affect the earliest born neurons, they are also likely to have lasting effects on the developing neonatal cortex.Because basic connectivity is established prenatally in most mammals, alterations to this scaffold could affect downstream steps of brain formation (77).Even transient changes have been shown to exert such an effect (78,79).Related to this, it would be important to assess whether the defects observed in our study could be rescued by overexpression of CNTNAP2 and at what point in development, rescue is no longer possible.It will also be critical to supplement the experiments in this study with CNTNAP2 null cultures derived from additional donor cell lines.Replicating our results in different genetic backgrounds will be a key next step.These findings could have important implications for the diagnosis and treatment of CNTNAP2-related disorders.They may also provide important insights for neurodevelopmental disorders more generally, including those not directly involving CNTNAP2, but that may be underpinned by similar pathophysiological mechanisms.

Loss of CNTNAP2
Loss of CNTNAP2 Alters Cortical Neuron Development Biological Psychiatry November 15, 2023; 94:780-791 www.sobp.org/journalLoss of Function of the Long Isoform of CNTNAP2 Does Not Impair Directed Differentiation of Human iPSCs to Cortical Neurons

Figure 1 .
Figure 1.Temporal expression of CNTNAP2 during early cerebral cortex development.(A) Diagram of the retinoid/dual-SMAD inhibition process used in this study for replaying cerebral cortex development from neural progenitor cells to excitatory cortical neuron differentiation and circuit formation.Examples of genes expressed in each stage of the process are shown in italics.(B) Efficient induction of cortical progenitor cells and cortical neurons from human iPSCs confirmed by the expression of key indicator genes in D30 progenitors and D50 to D80 neurons (n = 3 inductions).Pluripotency genes are downregulated, and progenitor/neuron-expressed genes are upregulated during the induction process, as measured with a custom Nanostring gene expression array (see Methods and Materials for details).(C) Diagram of known and predicted protein-coding isoforms of CNTNAP2: transcripts 201 (full-length), 203 (short), 205, and 207.Transcripts 201 and 203 are robustly expressed during cortical development.Notably, homozygous loss-offunction mutations predominantly overlap with the full-length isoform, as indicated (truncating mutations, red; microdeletions, blue).(D) The mRNA encoding the long isoform of CNTNAP2 increases in expression during neurogenesis and cortical neuron differentiation.Expression of the full-length form of CNTNAP2 was measured by quantitative reverse transcriptase polymerase chain reaction with primers targeting exons 3 and 4 and normalized to RAB7A (n = 3 inductions) (repeated-measures analysis of variance; post hoc paired t tests adjusted with Bonferroni correction).Expression was measured in iPSCs, cortical progenitor cells (D30) and 2 stages of neurogenesis and neuronal differentiation (D50 and D80).*p , .05;**p , .01;***p , .001.Mean expression at each time point is shown in green.(E) The full-length and short isoforms of CNTNAP2 are expressed in primary human fetal cortex and expression of full-length CNTNAP2 increases between 8 and 9 PCW.Isoform-specific relative expression of CNTNAP2 was derived from publicly available bulk RNA-Seq data of the human dorsolateral prefrontal cortex in the BrainSpan atlas.(F) CNTNAP2 protein levels increase over the period of cortical development from iPSCs, reflecting the temporal expression of the corresponding mRNA.Western blot of full-length CNTNAP2 protein and relative quantitation of CNTNAP2 levels normalized to b-actin are shown (n = 2 inductions) (same statistical tests as with quantitative reverse transcriptase polymerase chain reaction).*p , .05,**p , .01,***p , .001.Mean expression at each time point is shown in blue.Example blot from one induction (cell line = AD2.1) is provided.(G) CNTNAP2 is expressed as puncta on axons and dendrites of D50 human excitatory cortical neurons.Confocal images of CNTNAP2 immunostaining (red) appears along dendrites (MAP2, green), axons (tau, green) and cell bodies (nuclei labeled with DAPI, blue).By D75, CNTNAP2 localizes to synapses, as shown by the overlap with PSD-95 and Synaptophysin (both yellow).Scale bar = 5 mm.iPSC, induced PSC; mRNA, messenger RNA; PCW, postconception week; PSC, pluripotent stem cell; RNA-Seq, RNA sequencing; RPKM, reads per kilobase, per million mapped reads.

Figure 2 .
Figure 2. Loss of CNTNAP2 primarily alters gene expression in cortical neurons.(A) Single-cell RNA sequencing data of D50 forebrain cultures derived from CNTNAP2 null and isogenic control human induced pluripotent stem cells identified 6 major cell types (n = 2 null inductions, 2 control inductions, genetic background = NDC1.2).Data are represented as a UMAP projection of the 11,941 cells analyzed.(B) Expression of classifier genes used to assign cell type identities, including SOX2 for progenitor cells and DCX for postmitotic neurons.Cortical progenitor cells and cortical neurons were further identified by the expression of FOXG1 and EMX1 as well as TBR1, for neurons only.GAD2 identified inhibitory cells and RSPO2 identified medial telencephalon (cortical hem and choroid plexus).(C) Full-length CNTNAP2 is predominantly expressed by cortical neurons.Differential expression of CNTNAP2 was used as a proxy for full-length expression, which is specifically reduced in the null lines as shown in (D).The heatmap shows the log 2 FC between CNTNAP2 null samples and isogenic controls.Comparisons that were not significant (adjusted p value $ .05)are shown in gray.(D) Cortical neurons show the highest number of differentially expressed genes, signifying that loss of full-length CNTNAP2 in cortical neurons directly leads to altered gene expression.The number of genes shown includes only those passing statistical significance (adjusted p value , .05).(E) Cortical development is represented by a single trajectory in D50 forebrain cultures.Trajectory was inferred by pseudotime ordering of cortical progenitor cells and cortical neurons within the dataset.UMAP plot shows cells colored by relative pseudotime value with cells early in cortical development in purple and cells latest in development in yellow.Arrows indicate the trajectory among cortical progenitor cells (purple) and cortical neurons (green).(F) Expression of genes with known progenitor cell and neuronal expression in human cortical development validate the inferred trajectory.Relative gene expression in cells of the cortical lineage is shown ordered by cortical pseudotime.The line represents the fit of a negative binomial generalized additive model for the indicated gene along pseudotime separately for cells from CNTNAP2 null or isogenic control samples.Values from individual cells and fitted lines are colored by genotype.Genes regulating cortical progenitor cell maintenance (SOX2and PAX6) show the highest expression at the beginning of the trajectory, and genes known to increase with neuronal maturation (MAPT and DCX) show the highest expression toward the end.The expression patterns of these genes do not differ between CNTNAP2 null and isogenic control samples.Arrows under the x axes indicate the pseudotime values associated with cortical progenitors (purple) and cortical neurons (green).(G) Differentially expressed genes between CNTNAP2 null and control cortical neurons.The volcano plot shows the log 2 FC and associated p value for each gene, with genes passing statistical significance highlighted in red (adjusted p value , .05)(78 genes).Genes with a negative log 2 FC have higher expression in control neurons and those with a positive FC have higher expression in null neurons.(H) Differentially expressed genes between CNTNAP2 null and control cortical neurons show the highest expression toward the end of the cortical trajectory.Relative expression and generalized additive model-fitted lines for CALB2 and NTS (higher expression in null neurons) as well as LMO3, NEUROD2, and CSRP2 (higher expression in control neurons) is shown ordered by cortical pseudotime, as in (F).log 2 FC, log 2 fold change; prog., progenitor; PTh, prethalamus; UMAP, Uniform Manifold Approximation and Projection.

Figure 3 .
Figure 3. CNTNAP2 loss of function alters the production of cortical neuron subtypes.(A) Subclustering of cortical neurons reveals 4 subpopulations that are separated by relative maturity, as defined by increasing expression of neuronal maturation genes (MAPT and DCX) and decreasing expression of progenitor-associated genes (PAX6 and EOMES).UMAP projection shows cortical neurons colored by subpopulation.(B) The 2 genes most differentially expressed between CNTNAP2 null and isogenic control neurons, CALB2 and LMO3, separate late 1 and late 2 neuronal clusters, respectively, as do NTS and NEUROD2.UMAP projections of cortical neurons are colored by the relative expression of the indicated genes.(C, D)The relative proportion of the 2 most mature neuron subpopulations are shifted between CNTNAP2 null and control samples.In (C) cell density is shown overlaid onto the UMAP projection of cortical neurons (separated by genotype).The density of CALB2 1 , late 1 neurons is increased in the CNTNAP2 null samples, while the density of LMO3 1 , late 2 neurons is reduced.In (D) the percentage of all cortical neurons belonging to each subpopulation is plotted (separated by genotype).Error bars represent standard error of the mean.(E) Full-length CNTNAP2 is expressed in the most mature neuronal populations (late 1 and 2).Differential expression of CNTNAP2 was used as a proxy for full-length expression as in (C).Log 2 FC between CNTNAP2 null and control samples in the indicated population is shown as a heatmap with comparisons not passing statistical significance (adjusted p value $ .05) in gray.(F, G) Differential gene expression between CNTNAP2 null and control samples in late 1 (F) and late 2 (G) cortical neurons.Volcano plots show the log 2 FC and p value for each gene, with genes passing statistical significance (adjusted p value , .05)highlighted in red.In both late 1 and late 2 cortical neurons, CALB2 and NTS were highly upregulated in the CNTNAP2 null samples, while LMO3 was downregulated.Additionally, downregulation of multiple genes belonging to the cholesterol and fatty acid synthesis pathways, including SQLE, FDFT1, DHCR7, HMGCR, and HMGCS1, was also observed in CNTNAP2 null neurons from both late 1 and late 2 subtypes.Expression of example cholesterol and fatty acid synthesis genes in late 1 and late 2 neurons is shown as a dot plot separated by sample of origin.Dot color represents scaled expression levels and dot size the percentage of cells with nonzero expression.Ctrl, control; log 2 FC, log 2 fold change; UMAP, Uniform Manifold Approximation and Projection.

Figure 4 .
Figure 4. Loss of full-length CNTNAP2 reduces neurite branching and overall neuronal complexity in a cell autonomous manner.(A) CNTNAP2 null and isogenic control cultures were either plated into wells separately (i.e., by genotype) or mixed (cocultured).To facilitate the tracing of individual neurons, neurons of each genotype were nucleofected with different fluorescent reporters under the control of a neuron-specific promoter.For each mixed plating, 2 experiments were conducted with fluorescent proteins swapped between genotypes in the second iteration.(B) Three parameters of neurite outgrowth were measured: branch number, the sum of discrete segments composing a neuron; branch length, the distance from a branch point to a branch tip or to the next branch point; and total neuron length, the sum of all branch lengths.(C) Representative examples of labeled single CNTNAP2 null and control neurons used as input for neurite analysis.Scale bar = 200 mm.(D-F) CNTNAP2 null neurons have significantly reduced neuronal complexity.The number of branches (D), average branch length (E), and average total neuron length (F) were determined for CNTNAP2 null and control neurons from cultures either plated individually by genotype or mixed.CNTNAP2 null neurons showed significantly fewer branches per neuron, independent of plating type (D).Branch length (E) was significantly increased in mixed cultures only, and neuron length was significantly decreased in mixed cultures.*p , .05,**p , .01,***p , .001;repeated-measures analysis of variance; post hoc estimated marginal means analysis with Bonferroni correction; n = 648 control neurons, 662 null neurons from 2 inductions per genotype; error bars represent 95% confidence intervals.Both means (large dots) and individual data points (small dots) are plotted.CaMKII, calcium/calmodulin-dependent protein kinase II; Ctrl, control.

Figure 5 .
Figure 5. Loss of full-length CNTNAP2 alters developing neural circuit activity.(A) Overview of the experimental setup to study network-level neuronal activity in induced pluripotent stem cells-derived CNTNAP2 null and isogenic control cultures.To record activity, neuronal cultures were either infected with a lentiviral vector expressing a genetically encoded orange, fluorescent calcium indicator (mRuby based) or plated onto an MEA plate.Culture activity was then longitudinally measured by live calcium imaging or MEA over a course of w30 to 60 days.(B) Cortical neurons are efficiently generated from CNTNAP2 null induced pluripotent stem cells, as demonstrated by confocal imaging of immunostained D50 isogenic control and CNTNAP2 null cortical differentiations for that of control cultures (Sidak's multiple comparisons test; **p , .01,***p , .001,****p , .0001).Error bars show standard error, with each data point representing the average of 8 wells per genotype (n = 2 control inductions, 2 CNTNAP2 null inductions).(D) Trends similar to those observed in the calcium imaging results were observed in the MEA experiments.D35 control and CNTNAP2 null cortical cultures were plated on MEAs, and activity was recorded every 24 to 48 hours.CNTNAP2 null cultures had significantly more spikes per active electrode at (D60, D72, and D77) and a higher burst frequency at D72 and D77 (pairwise t tests; *p , .05,**p , .01;n = 1 control induction, 1 CNTNAP2 null induction).Both parameters then decreased in the null samples while continuing to increase in controls.Error bars show standard error.Ctrl, control; MEA, multielectrode array.Loss of CNTNAP2 Alters Cortical Neuron Development Biological Psychiatry November 15, 2023; 94:780-791 www.sobp.org/journalCNTNAP2 null cultures, dropping below the level of controls by D81.
Loss of CNTNAP2 Alters Cortical Neuron Development Biological Psychiatry November 15, 2023; 94:780-791 www.sobp.org/journal785 CNTNAP2 Loss of Function Alters Neuronal Activity in Developing Human Cortical Networks Given the reduction observed in neurite branching, we measured neuronal activity to investigate whether this affected cortical neural network development.To measure networklevel neuronal activity, we used 2 complementary but significant effects of culture age and CNTNAP2 LOF on both the number of active objects and the average burst rate (Figure 5C) (active objects: F 39,560 = 297.4,p , .0001;burst rate: F 39,560 = 68.21,p , .0001).Notably, several other activity parameters also showed similar trends (Figure Alters Cortical Neuron Development 788Biological Psychiatry November 15, 2023; 94:780-791 www.sobp.org/journalBiological Psychiatry