Cell-type specific global reprogramming of the transcriptome and epigenome in induced neurons with the 16p11.2 neuropsychiatric CNVs

Abstract

Copy number variants (CNVs), either deletions or duplications, at the 16p11.2 locus in the human genome are known to increase the risk for autism spectrum disorders (ASD), schizophrenia, and several other developmental conditions. Here, we investigate the global effects on gene expression and DNA methylation using an induced pluripotent stem cell (iPSC) to induced neuron (iN) cell model system derived from 16p11.2 CNV patients and controls. This approach revealed genome-wide and cell-type specific alterations to both gene expression and DNA methylation patterns and also yielded specific leads on genes potentially contributing to some of the phenotypes in 16p11.2 patients. There is global reprogramming of both the transcriptome and the DNA methylome. We observe sets of differentially expressed genes and differentially methylated regions, respectively, that are localized genome wide and that are shared, and with changes in the same direction, between the deletion and duplication genotypes. The gene PCSK9 is identified as a possible contributing factor to symptoms seen in carriers of the 16p11.2 CNVs. The protocadherin (PCDH) gene family is found to have altered DNA methylation patterns in the CNV patient samples. The iPSC lines used for this study are available through a repository as a resource for research into the molecular etiology of the clinical phenotypes of 16p11.2 CNVs and into that of neuropsychiatric and neurodevelopmental disorders in general.

Introduction

Neuropsychiatric disorders have a high prevalence in humans, ranging from ~.57% for schizophrenia to 19.1% for any anxiety disorder in the US based on reports from the National Institute of Mental Health (NIMH). All of these disorders are estimated to have high levels of heritability. While the specific underlying molecular and cellular mechanisms are still mostly unknown, multiple regions of the genome as well as increasing numbers of individual candidate genes have been found to be associated with these disorders. Several high-confidence candidate loci now exist and those with the highest penetrance are large copy number variants (CNVs) [1, 2].

CNVs are large deletions or duplications in the human genome that are at least 1 kb in size but often range into the hundreds of thousands to millions of base pairs. There are around 70 to 120 CNV regions that are associated with genomic disorders [3]. Two large CNVs with the strongest association to neuropsychiatric, and generally neurological, disorders are the 16p11.2 BP4-BP5 deletion CNV (MIM 611913) and the duplication CNV (MIM 614671) in the same locus, which are known to be present in about 1% of ASD patients [4]. Furthermore, the duplication CNV in this locus increases the risk for developing schizophrenia by 14-fold [5]. The 600kb CNV can manifest as either a deletion or a duplication, with start and endpoints within the same segmental duplications (SegDups), and is found in at least 3/10,000 individuals [4, 6]. Along with the neuropsychiatric disorders, the 16p11.2 CNV has been associated with several other developmental phenotypes including speech and language delay, seizures, intellectual disability (ID), developmental delay (DD), developmental coordination disorder (DCD), anxiety, and attention-deficit/hyperactivity disorder (ADHD) [1, 6–10]. Additionally, the deletion and duplication cases are known to often show mirrored phenotypes for head circumference and body mass index (BMI) [11]. Specifically, the deletion patients will typically develop macrocephaly, whereas the duplication patients tend to develop microcephaly [9]. This trend has some variance, with 10% of duplication carriers found to have microcephaly and 17% of deletion carriers found to have macrocephaly [9]. For BMI, deletion carriers show an increase in BMI by age 7 resulting in ~75% of adult carriers being obese. Conversely, duplication carriers are at an eightfold risk of being underweight [12].

Previous studies of the molecular and cellular impact of these CNVs have already yielded important insights but also have left many fundamental questions unanswered. The problem is particularly challenging given the large degree of phenotypic diversity and variance in the patients, the large size of the genomic region affected and the concurrent large number of genes that are encoded within that region. On top of the variance and noise generated in the data from the region, there is also a lack of primary human neuronal sample tissue to perform these studies in, while limitations exist in previously used model systems. For instance, achievements gained by employing human lymphoblastoid cell lines (LCLs) derived from patients need further validation in cell types closer in identity to those found in the brain. In the meanwhile, notable discrepancies lie between 16p11.2 mouse model systems and human neuronal models since the gene content of the (engineered) CNV in the mouse genome and the phenotypic level are not entirely identical to that in the human patients [13–15].

Recently, additional important work has emerged, namely induced pluripotent stem cells (iPSC) have been used to engineer the 16p11.2 CNV into a control cell line [16–18], and contrasting cellular phenotypes have been described in 2D cultures in neurons derived from patient iPSCs [11, 19]. The iPSC-iN system provides an unrivaled opportunity to dissect early neurodevelopmental dysfunction associated with 16p11.2 CNVs and shed light on the molecular etiology of multiple psychiatric disorders. For instance, induced excitatory neurons from 16p11.2 duplication carriers with SCZ showed deficits in network synchrony, dendrite outgrowth, and calcium handling [18], while iPSC-derived dopaminergic neuron harboring CNVs of 16p11.2 deletion displayed hyperactive networks [17]. Furthermore, cortical organoids grown from iPSCs with either the deletion or the duplication CNV were shown to be recapitulating the macrocephaly and microcephaly phenotypes as well as again exhibiting contrasting cellular phenotypes, in addition to transcriptomic deregulation of pathways relevant to neuronal development and function [13, 20].

Here we perform a study on the alterations to gene expression and DNA methylation while 16p11.2 CNV patient iPS cell lines are induced into a neuronal cell type. Using an available resource for 16p11.2 CNV patient iPSC cell lines from the Simons Foundation Autism Research Initiative (SFARI), we were able to evaluate the molecular impact of the 16p11.2 CNV and identify potential key players as well as unexpected potential mechanisms in the molecular etiology of some of the clinical phenotypes seen in patients.

Materials and Methods

See detailed descriptions of Materials and Methods in Supplementary Materials.

Results

16p11.2 CNV patient-derived iPSCs form the basis of iNs

To examine the molecular effects of 16p11.2 CNVs in a biologically relevant neurological tissue, we induced 15 iPSC lines (4 Controls, 6 Deletions, 5 Duplications) into iNs using a single-step induction approach [21] (Figure S1A). Low-coverage whole-genome sequencing was performed on the iPSCs to confirm the presence and the boundaries (hg38, chr16: 29,570,001–30,190,000) of CNVs in the patient lines, visualized in Integrative Genomics Viewer (IGV) [22] (Figure S1B). No other large CNVs were observed in any iPSC lines. The copy number of the 16p11.2 region was further verified through ddPCR using TaqMan copy number assays (Figure S2). GFP imaging and staining of neuronal cell markers TUB and MAP were performed to confirm that the differentiated cells were neuronal cells (Figure S1C). The differentiation process of iPSC cell lines into iNs was also verified by normalized expression levels from RNA-Seq of canonical iPSC-specific and neuron-specific genes (Figure S3). Pathway analysis of the upregulated genes in the iNs compared to the iPSCs confirmed that multiple neuronal pathways were activated (Figure S4). The iPSC-iN model system makes it possible to study the molecular effects of the 16p11.2 CNVs in the cell type, i.e. neurons, which is essential to psychiatric disorders, in an efficient and well-controlled manner. By simulating the differentiation process from pluripotent stem cells to early-stage neurons, in a simplified manner, this system could provide information on some of the developmental stages during which it could be speculated that the foundations for some of the patient phenotypes perhaps had been laid, and promises to give insights into the etiology of psychiatric disorders associated with 16p11.2 CNVs.

Gene expression levels in patients change within the CNV region as expected

The principal component analysis (PCA) based on the RNA-Seq data showed the distance among cell lines at the iPSC stage and iN stage (Figure S5) and helped exclude three outlier cell lines at the iN stage for the following analysis. We observed expression of the same 21 genes within the 16p11.2 region in iPSCs and iNs. The most likely scenario regarding gene expression is that the deletion of one copy of a gene will result in a decrease in expression level while its duplication will result in an increase compared with the controls. In line with this expectation, we found that all the expressed genes within the CNV region exhibited a change in gene expression that followed the change in genomic copy number, at both the iPSC and iN stages (Figure 1A and 1B). We did not observe evidence of dosage compensation of expression for any genes within the CNV region in patients. Differential expression analysis between patients and controls revealed that the most significant differentially expressed genes (DEGs) genome-wide were located within the 16p11.2 region in both Duplications vs. Controls and Deletions vs. Controls comparisons at both cell type stages (Figure 1C–F), indicating that the effects of the 16p11.2 CNV on the genes directly associated via their genomic position are the strongest.

Figure 1. Gene expression is altered within the CNV region and genome-wide.

Log2 fold change and 95% confidence interval of the genes within the CNV region and its flanking regions are shown in (A) for iPSCs and in (B) for iNs. The comparison of Deletions vs. Controls is shown in red and Duplications vs. Controls in blue. −log10(p-values) of differential expression (DE) analysis across the genome are shown in (C) for Deletions vs. Controls and in (D) for Duplications vs. Controls for iPSCs, in (E) for Deletions vs. Controls and in (F) for Duplications vs. Controls for iNs. Genome-wide significance is based on FDR < 0.05 for iPSCs and < 0.1 for iNs indicated by the horizontal lines. Volcano plots of DE analysis are presented in (G) for Deletions vs. Controls and in (H) for Duplications vs. Controls for iPSCs, in (I) for Deletions vs. Controls and in (J) for Duplications vs. Controls for iNs. Significance is based on FDR < 0.05 for iPSCs and < 0.1 for iNs indicated by the horizontal dashed lines and 1.5 fold change of expression levels is denoted by the vertical dashed lines.

Gene expression is altered genome-wide

Gene expression alterations were not localized to only the CNV region. DEGs were found genome-wide for both CNV types at both cell stages (Figure 1C–F). At the iPSC stage, there were 12,291 genes with detectable expression genome-wide. Of those genes, 183 were identified as DEGs in the comparison of Deletions vs. Controls with 50 of those being up-regulated and 133 down-regulated (at FDR < 0.05; Figure 1G and 2A). In the comparison of Duplications vs. Controls, there were 1,809 DEG, with 500 up-regulated and 1309 down-regulated genes (at FDR < 0.05; Figure 1H and 2A). At the iN stage, there were 11,922 genes with detectable expression. The Deletions vs. Controls analysis detected 90 DEGs at FDR < 0.1 (39 up-regulated, 51 down-regulated) (Figure 1I and 2B) while the Duplications vs. Controls analysis identified 830 DEGs at FDR < 0.1 (541 up-regulated, 289 down-regulated)(Figure 1J and 2B).

Figure 2. Shared DEGs between comparisons.

Venn diagrams show the number of DEGs genome-wide from each of the four analyses and the number of DEGs that overlap at the iPSC stage (A) and the iN stage (B). Shared DEGs between the comparison of Deletions vs. Controls and Duplications vs. Controls in iPSCs by the pipeline DESeq2 (C) and by Cuffdiff2 (E). Shared DEGs between the comparison of Deletions vs. Controls and Duplications vs. Controls in iNs by DESeq2 (D) and by Cuffdiff2 (F). Log2 fold change and 95% confidence interval of the genes are shown. Dashed black circle represents log2 fold change = 0. In panel (E) the gene PCSK9, the only gene in that comparison that does not follow the pattern of same-direction change of gene expression levels, is indicated by a green arrow.

Most of the overlapping DEGs between Deletions and Duplications outside the 16p11.2 region show the same direction of change in iPSCs and iNs

Given that the deletion and duplication in the patients impact the same genes in the 16p11.2 region, we speculated that there might be shared expression network effects. Analyzing the gene expression data for such effects, we did indeed observe overlapping DEGs between the comparison of Deletions vs. Controls and Duplications vs. Controls at both cell stages. In iPSCs, there were 80 overlapping DEGs between the two comparisons, 72 of which were outside the 16p11.2 region. Strikingly, these DEGs showed a strong trend towards having the same direction of fold changes between Deletions and Duplication DEGs, even for DEGs that are located outside the CNV boundaries, and in fact genome-wide (Figure 2A and 2C). In iNs, there were 37 overlapping DEGs between the two comparisons, 19 of which were outside the 16p11.2 region. Again the majority of these genes (17 of 19) displayed expression changes into same direction (Figure 2B and 2D).

Two alternative RNA-Seq pipelines employed for confirmation purposes (Tophat2 + Cuffdiff2; limma 3.42.2) generated a similar pattern of consistent fold changes for the overlapping DEGs between Deletions vs. Controls and Duplications vs. Controls at both cell stages (Figure 2E and 2F; Figure S6). Moreover, to validate our finding in independent sets of data, we re-analyzed the RNA-Seq data of both human LCLs with 16p11.2 CNV and comparable mouse brain tissues from the study of Blumenthal et al. [14], using the same pipeline as for our own RNA-Seq data. The results of reanalysis showed a similar same-direction fold-change pattern (Figure S7–S9, Table S2–S3, details in Supplementary Materials).

Overlapping DEGs at the iN stage are relevant for neuronal function

We carried out pathway enrichment analysis on the overlapping DEGs between Deletions vs. Controls and Duplications vs. Controls at the iN stage. The top pathway enriched in these 19 shared DEGs outside the 16p11.2 region was ‘neuron fate commitment’ (p-value 0.015). Seven of these DEGs were associated with neuronal functions: ST3GAL3, OLIG1, OLIG2, D2HGDH, SLC25A1, FBLN1 and NEUROG1. Mutations in ST3GAL3 may play a role in a form of nonsyndromic cognitive disability [23–25]. OLIG1 and OLIG2 are relevant for neural crest differentiation and neural stem cell differentiation pathways [26, 27]. OLIG2 is required for oligodendrocyte differentiation and the development of somatic motor neurons in the hindbrain [27, 28]. It also cooperates with OLIG1 to play a role in the development of the embryonic neural tube [29]. Mutations in D2HGDH and SLC25A1 have been associated with a rare recessive neurometabolic disorder D-2-hydroxyglutaric aciduria [30, 31]. FBLN1 is related to developmental delay-central nervous system anomaly-syndactyly syndrome [32]. NEUROG1 is a transcriptional regulator and is involved in the initiation of neuronal differentiation [33, 34].

PCSK9 knockdown leads to altered developmental phenotypes in zebrafish

One of our analysis pipelines (Tophat2 + Cuffdiff2) identified the gene PCSK9, located outside the 16p11.2 region, as the only shared DEG between Deletions vs. Controls and Duplications vs. Controls that is exhibiting expression into opposite directions, in iPSCs (Figure 2E). At the iPSC stage, PCSK9 was found to be significantly up-regulated in the Deletions (log2 fold change 1.10) but significantly down-regulated in the Duplications (log2 fold change −1.16). However, at the iN stage, PCSK9 was significantly up-regulated (log2 fold change 1.72) in the Duplications and, while it is not a DEG in the comparison of Deletions vs. Controls, it was also up-regulated (log2 fold change 1.42) in the Deletions. Gene expression changes were verified using qPCR (Table S4).

To test whether mirror-image changes in PCSK9 expression alone may lead to developmental phenotypes that could be associated with those seen in 16p11.2 patients we employed the zebrafish model. We observed no detectable changes in the control (with the control morpholino (ConMo) injected) and PCSK9 mRNA induced overexpression zebrafish while phenotypic alterations were observed in the PCSK9 morpholino knockdown zebrafish. The knockdown resulted in effects on embryonic and early brain development such that the zebrafish exhibited a marked curvature of the spine and a darkening of and reduction of brain tissue (Figure 3A1). Additionally, these zebrafish had a reduced body length (3.85 ± 0.0016mm (ConMO) vs. 2.64 ± 0.0057mm (PCSK9MO) vs. 3.77 ± 0.0012mm (PCSK9 mRNA); p <0.05, Student’s t-test, N=3 independent tests) and an increased interocular distance (0.12 ± 0.00010mm (Con MO) vs. 0.40 ± 0.0036mm (PCSK9MO) vs. 0.12 ± 0.00010mm (PCSK9 mRNA); p <0.05, Student’s t-test, N=3 independent tests) resulting from neural degeneration and oedema when compared to the control and overexpression zebrafish (Figure S10). Caspase-3 immunostaining performed at 1 day post fertilization (dpf) and 5 dpf showed an increase in apoptosis in PCSK9 knockdown zebrafish when compared to the other two groups with more caspase-3 fluorescence at 1 dpf (Figure 3A2).

Figure 3. Reduction of PCSK9 expression in zebrafish impacts development.

(A) PCSK9 expression was manipulated in zebrafish by injecting a PCSK9 morpholino (PCSK9MO) to knockdown expression and PCSK9 mRNA to increase expression to comparable levels seen in the 16p11.2 CNV patients. A control morpholino (Con MO) was also injected to ensure that none of the observed effects resulted from morpholino injection. (A1) The knockdown resulted in defects on early brain/embryonic development; primarily the zebrafish had a curvature of the spine and a darkening and reduction of the brain tissue. By comparison, the control and overexpression zebrafish were unaffected. (A2) Caspase-3 immunostaining revealed that an increase in apoptosis was occurring in the 1dpf PCSK9 knockdown zebrafish within the hindbrain (marked by dotted line). (B) At 5 dpf, Alcian Blue staining shows disruption of the ventral head structures including the jaw, presumptive operculum and pectoral fins of the PCSK9 −/− zebrafish which was not seen in the in-clutch wildtype zebrafish.

The phenotype of heterozygous mutant crosses that contained PCSK9 +/+ PCSK9 +/− and PCSK9 −/− fish was characterized from 1 dpf through 5 dpf. While there were imperceptible differences between wildtype and heterozygous mutants, we detected at appropriate Mendelian ratios that PCSK9 −/− mutant fish (verified post-hoc by larval genotyping) exhibited similar developmental phenotypes in body plan and interocular distance as seen with morpholino knockdown, indicating the specificity of the aberrations with PCSK9 perturbation. The effect of null mutation was embryonic lethal, fish with this genotype did not survive past 5–6 dpf.

Prior to lethality, we detected a disruption of the development of ventral structures and further characterized these with Alcian Blue staining. Following staining of the cartilage at 5 dpf, we observed abnormalities of the ventral head structures including the jaw, presumptive operculum and pectoral fins in the PCSK9 −/− zebrafish, which was not seen in the in-clutch wildtype zebrafish (Figure 3B). These results suggest that part of PCSK9’s function could be involved in the specification of ventral structures, failure of which, in complete absence, affect overall development with considerable consequences to lifespan.

DNA methylation regulation is altered genome-wide

Gene expression is controlled by epigenomic processes such as DNA methylation, and chromatin changes have been implicated in the molecular etiology of disorders such as ASD and schizophrenia [35, 36]. After profiling the genome-wide DNA methylation patterns in the CNV lines and controls, we found similar numbers of autosomal differentially methylated regions (DMRs) between Deletions vs. Controls and between Duplications vs. Controls at both the iPSC and iN stages (Figure 4). In iPSCs, we found 211 DMRs in the comparison of Deletions vs. Controls (148 hypermethylated and 63 hypomethylated in the Deletions) and 154 DMRs in Duplications vs. Controls (109 hypermethylated and 45 hypomethylated in the Duplications) (Figure 4A, 4C and 4D). In iNs, there were 191 DMRs in Deletions vs. Controls (150 hypermethylated and 41 hypomethylated in the Deletions) and 218 DMRs in Duplications vs. Controls (195 hypermethylated and 23 hypomethylated in the Duplications) (Figure 4A, 4E and 4F).

Figure 4. Analysis of differential DNA methylation between patients and controls.

(A) The total number of differentially methylated regions (DMRs) found in each analysis and the breakdown of DMRs found to be hypermethylated or hypomethylated compared to controls. (B) The number of unique genes with at least one DMR located within the gene’s promoter region or gene body. −log10 (p-values) across the genome are shown in (C) for Deletions vs. Controls and in (D) for Duplications vs. Controls for iPSCs, in (E) for Deletions vs. Controls and in (F) for Duplications vs. Controls for iNs. Genome-wide significance is based on FDR < 0.05 indicated by the horizontal black lines. (G-J) Overlapping DMRs from cell stage and CNV-type comparisons plotted by methylation value and labeled by a unique number of value and chromosome location. Plots shown are DMRs overlapping at the iPSC stage (G), at the iN cell stage (H), between cell stages for deletion patients (I), and between cell stages for duplication patients (J). Dotted line represents zero change in methylation.

In all four comparisons, the majority of the DMRs were found to be hypermethylated in the CNV lines. Compared with all the methylated regions genome-wide, the hypermethylated regions was significantly enriched in the DMRs called between the CNV lines and controls in each comparison (Fisher’s exact test p-value 5.83E-4 in Deletions and 7.80E-4 in Duplications for the iPSCs; 1.02E-14 in Deletions and < 2.20E–16 in Duplications for the iNs).

When examining the genomic locations of the DMRs we found that, in line with the gene expression changes, the epigenetic alterations were spread across the genome and were not exclusively localized to chromosome 16 or enriched in the 16p11.2 CNV locus. In fact, no DMRs were observed in any comparisons within the 16p11.2 CNV region, indicating that the presence of the CNVs does not impact the DNA methylation patterns of the genes in this region.

DMRs overlapping between Deletions and Duplications display same-direction DNA methylation changes

Similar to the gene expression analyses, we asked whether DMRs from the different comparisons overlapped and whether there was a tendency of same-direction change for the shared DMRs between Deletions and Duplications. We identified 62 overlapping DMRs between Deletions vs. Controls and Duplications vs. Controls in iPSCs (Figure 4G). Intriguingly, all but two of these overlapping DMRs exhibited DNA methylation changes into the same direction in both Deletions and Duplications compared with Controls. There were 70 overlapping DMRs between Deletions vs. Controls and Duplications vs. Controls in iNs and all of them showed same-direction DNA methylation change (Figure 4H).

We also analyzed DMRs between cell stages. For the comparison of Deletions vs. Controls, there were 146 overlapping DMRs between iPSCs and iNs (Figure 4I). For the comparison of Duplications vs. Controls, there were 108 overlapping DMRs between the two cell differentiation stages (Figure 4J). The directions of DNA methylation changes for these overlapping DMRs were also consistent between these two cell differentiation stages for both Deletions vs. Controls and Duplications vs. Controls, indicating that many of the DNA methylation changes associated with the 16p11.2 CNVs in iNs already occurred at the iPSC stage.

DNA methylation changes and gene expression changes are complementary rather than overlapping

We assigned the DMRs to their closest transcription start sites (TSSs) of the protein coding genes. At the iPSC stage, DMRs were assigned to 158 genes in Deletions vs. Controls and 125 genes in Duplications vs. Controls. Interestingly, 11 of the genes are members of the PCDH family (Figure 4B). Among them, 86 in Deletions vs. Controls and 64 in Duplications vs. Controls were expressed in iPSCs respectively. However, only 3 genes were DEGs in Deletions vs. Controls and 11 were DEGs in Duplications vs. Controls. For Deletions vs. Controls, all the DMRs assigned to the 3 DEGs were hypermethylated in Deletions and 2 of the genes were downregulated in Deletions. For Duplications vs. Controls, 10 of the DMRs assigned to the 11 DEGs were hypermethylated in Duplications but only 4 of the associated DEGs were downregulated in Duplications.

At the iN stage, DMRs were assigned to 145 genes in Deletions vs. Controls and 163 genes in Duplications vs. Controls. Twelve of the genes are members of the PCDH family (Figure 4B). We observed expression of 79 genes in Deletions vs. Controls and 84 in Duplications vs. Controls in iNs. Only one gene (TSTD1) was a DEG in Deletions vs. Controls and 6 genes were DEGs in Duplications vs. Controls. Gene TSTD1 was hypermethylated and downregulated in Deletions. All the DMRs assigned to the 6 DEGs in Duplications vs. Controls were hypermethylated in Duplications and 4 of the DEGs were downregulated in Duplications. The lack of overlap between differentially methylated genes and DEGs in both iPSCs and iNs indicates that the DNA methylation changes associated with the 16p11.2 CNVs are not directly reflected by the expression changes. Taken together, our results suggest that the DNA methylation changes and gene expression changes in the 16p11.2 CNV lines were complementary rather than overlapping.

Discussion

We demonstrate here that the 16p11.2 CNV iPSCs that are available from the SFARI foundation can be directly induced into early-stage neurons. Such iNs are an efficient option for obtaining sufficient numbers of physiologically relevant cells in order to perform multiple functional genomics assays. In our study we assayed the iNs at an early stage. However, with additional investment of time and resources, it is also possible to obtain mature iNs by growing them on mouse glia. Such mature iNs are a well-established tool for neurodevelopmental and neurophysiological analyses [37].

The large CNVs on chromosome 16p11.2 generally had the expected effect on expression levels for the genes that are encoded within the CNV boundaries, namely the expression changes follow the changes in gene copy number, as reported by other recent studies employing human iPSC-derived models [11, 13, 17]. In addition, we observed that there is a marked genome-wide effect on gene expression levels, well beyond the boundaries of the CNV or even chromosome 16 as a whole, as had been seen in other model systems for the 16p11.2 CNVs [11, 13, 17, 18, 20]. There are two genes encoded within the CNV boundaries that act as transcription factors, MAZ and TBX6 [38, 39]. This differential expression of two transcription factors could be the basis for at least some of the genome-wide expression level changes that we observed. However, an analysis of transcription factor binding motifs did not uncover an enrichment for the known binding motifs for MAZ and TBX6 (data not shown), leaving this possible explanation for some of the global gene expression changes unconfirmed at this time. Many of the genes encoded within the CNVs can be expected to play potentially important roles in functional pathways, which could then also lead to other genes in such pathways to exhibit altered levels of expression.

When we explored whether there are DEGs or DMRs, that overlap between Deletion and Duplication CNVs, we made the unexpected observation that such overlapping DEGs and DMRs doexist, and these overlapping DEGs and DMRs almost exclusively have the same direction of change under both deletion and duplication conditions. This opens up the possibility that there is a molecular regulatory mechanism that is affected by both the deletion and the duplication genotype in a copy-number independent manner and subsequently acts on a set of ‘core’ genes and regulatory regions. Furthermore, regarding the potential existence of this core set of overlapping DEGs and DMRs in 16p11.2 syndrome, one could speculate that these overlapping genes and regulatory regions may be associated with, on the molecular level, the phenotypes that are shared in both deletion and duplication carriers, while the non-overlapping DEGs and DMRs might be to a greater degree involved in the molecular basis of the divergent phenotypes seen in either deletion or duplication patients. Our reanalysis of the gene expression data from the Blumenthal et al. [14] study that used both the human LCLs and mouse model for both the deletion and duplication CNVs revealed a similar finding of overlapping non-CNV DEGs in the 16p11.2 human LCLs and mouse model changing expression in the same direction in both the deletion and duplication genotypes, lending further support to our observation in the human iPS-to-iN model. The identified enrichments in biological processes involving synaptic signaling, as well as neuronal components involving dendrites and somatodendritic compartments are in line with the findings of previous studies [18, 19], reporting morphological changes in soma size, dendritic length and synaptic density in the differentiated neurons of 16p11.2 CNV patients, as well as with the findings in cortical organoids with the 16p11.2 CNVs [13, 20].

Our observation of opposite-direction gene expression level changes for PCSK9 allows for speculation on a potential role of this gene in the etiology of the opposing phenotypes of the deletion and duplication patients. One might hypothesize that PCSK9 could be contributing to both the neurological impairment [40] and body-mass-index phenotypes [41], also considering the reduction in brain mass and body length seen in some of the zebrafish experiments. Such reductions were also consistent to those seen before with PCSK9 knockdown in zebrafish [42]. Our data also indicate that changes in PCSK9 expression levels at critical stages of development could provide insights into its role in the development of facial abnormalities seen in 16p11.2 patients, for example the observed jaw malformations. In our zebrafish experiments, we observed that a reduction of PCSK9 expression levels resulted in major disorganization and underdevelopment of ventral jaw structures. One needs to bear in mind, however, that PCSK9 in humans may play a smaller role in the development of central nervous system than the large role for this gene in this context that is established in zebrafish [42, 43], and PCSK9 knockout mouse models do not exhibit facial or head circumference abnormalities, as reported in previous studies [43, 44]. In this context, much further work is needed.

The CNVs on 16p11.2 alter DNA methylation patterns similarly in both deletion and duplication patients as well as at both differentiation stages. Overall, the CNVs increase levels of DNA methylation across the genome except within the CNV region itself, where there is little to no significant change in DNA methylation levels. Chromosome 16 as a whole is also not disproportionately impacted on the level of DNA methylation. As there is generally a similar genome-wide impact on DNA methylation in all four analyses, it could be that the deletion and duplication of the 16p11.2 region are affecting the same or similar global DNA methylation regulation pathways.

We did not find a strong correlation between the altered DNA methylation patterns and specific changes in gene expression as far as the individual affected loci are concerned. However, the PCDH gene family stood out as a candidate for follow-up studies. Dysregulation of the protocadherins on both the levels of epigenetic marks and gene expression has been found to be associated with human disease, including neurological disorders. The review by El Hajj et al. [45] outlines the known associations that these dysregulations have with disorders such as schizophrenia [46, 47], bipolar disorder [47], and ASD [48]. We did not observe altered expression for the PCDH genes. At the same time, PCDH expression levels overall were very low in the early neuronal cell type that is being replicated by the iN cells in this study. One interpretation of these observations could be that differential DNA methylation simply does not always lead to a direct effect on gene expression, i.e. a DMR does not strictly affect the expression of the nearest gene, but potentially has an impact on genes over longer distances. The complicated relationship between the DNA methylation and gene expression has been reported and discussed in the literature [49, 50]. Some DNA methylation changes may carry over across several cell types during development, with other factors (e.g. chromatin remodelers and transcription factors) involved, following rules that are still only very incompletely understood. Although it may not be possible to directly tie individual DMRs to individual DEGs, the finding of DMRs in 16p11 CNV cell models reported here still remains for the first time that both the 16p11 deletion and the 16p11 duplication can be associated with genome wide, non-random, alterations in DNA methylation patterns.

Data availability

The data that support the findings of this study have been deposited in NCBI with the accession code GSE133615 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE133615).

Code availability

Analysis codes will be made available to any interested reader.