Genome-wide and single gene DNA methylation patterns of isogenic hESC- and iPSC-derived neurons are very similar
The current study addressed the question whether iPSC-derived neurons display similar DNA methylation patterns compared to hESC-derived neurons. As a first step, we compared isogenic hESC- and iPSC-derived neurons, which allow the analysis of DNA methylation differences independently of individual genetic variations. To that end, we already differentiated thoroughly the characterized and stable populations of hESC-derived stable neural stem cells (NSCs) from the female hESC line I3 into 6-week-old neurons (Fig. 1). These hESC-derived NSCs, also referred to as long-term self-renewing neuroepithelial stem cells (lt-NES cells) differentiate predominantly into GABAergic neurons with a posterior identity corresponding to the ventral hindbrain area [14]. We reprogrammed these NSCs to iPSC and subsequently differentiated them into three clonal populations of iPSC-derived neurons, which were compared with neurons generated from the parental hESC (Fig. 1). A genome-wide analysis was carried out using Illumina 450 K bead arrays measuring DNA methylation levels at > 450.000 CpG sites covering promoters and putative regulatory domains of all designable RefSeq genes.
A Pearson correlation analysis of gene-linked CpG methylation revealed at least a 96% concordance of iPS-NSC and iPS-Neurons compared to their hESC-derived counterparts (Fig. 2a). The interclonal variance was small; NSC and neurons derived from the three different iPSC clones showed a concordance for DNA methylation levels of at least 97% (Fig. 2a).
Differentially methylated CpGs (DMCGs) in iPSC-derived neurons compared to their hESC-derived counterparts (defined by a methylation difference > 20%) were analyzed in more detail; 8.680 (42.35%) of these DMCGs were hypomethylated, and 11.816 (57.65%) were hypermethylated in iPS- compared to hES-Neurons (Fig. 2b). Transcription start sites were not overrepresented in these DMCGs compared to the overall 450 K array annotation (Fig. 2c upper panels). More hypomethylated CpGs were found in CpG islands (CGI) and more hypermethylated CpGs in non-CGI DNA sequences (Fig. 2c lower panels). We also analyzed the standard deviation (> 0.2) of single DMCG in iPS-NSC (Fig. 2d) and iPS-Neurons (Fig. 2e) to identify an interclonal variation of DMCG. We identified only 3.051 and 3.264 interclonal DMCG sites with a standard deviation of > 0.2 for iPS-NSC and iPS-Neurons, respectively. We found a high correlation of variation between iPSC-derived NSC and neurons (Fig. 2d–f). To identify gene loci showing the most prominent alterations, we carried out an interindividual analysis (hES-Neurons vs. each iPSC-derived neuronal clone) applying a more stringent threshold of > 0.5 methylation difference (Additional file 1 A–C). In line with the previous results, we could identify only a few DMCGs (n = 1010 up to n = 1802) being differentially methylated comparing hES-Neurons to all three iPSC-derived neuronal clones (Additional file 1 A–C). The data indicate that (i) different iPSC neuronal clones exhibit almost the same amount of DMCG when compared to their hESC counterparts and (ii) that the differences are not biased toward certain genomic sites, keeping in mind that the distribution of CpGs on the 450 K array across the genome is not equally distributed per se. Using Fisher’s test, we could, however, identify a highly significant variation for methylation differences on the X chromosome in general (Additional file 1 D).
Consistent gene expression patterns of isogenic hESC- and iPSC-derived neurons
To compare gene expression patterns of hESC- and iPSC-derived neurons, we performed HT12v4 array experiments covering more than 47.000 transcripts. We tested the technical variation of the assay by hybridizing all the neuronal samples in duplicates. We observed high fidelity illustrated by correlation values of greater than 98% between each pair of replicates and a high interclonal correlation of hESC- and iPSC-derived neurons (Additional file 2 A–D). For all downstream comparisons, replicate 1 of each sample was used.
Pearson correlation analysis revealed at least a 96% concordance of gene expression profiles of iPSC- and hESC-derived neurons; interclonal variance was minimal (< 4%, Fig. 3a, Additional file 2 E, F). These data revealed that neurons generated from iPSC are very similar to their hESC-derived counterparts at the gene expression level. Hierarchical cluster analysis revealed that hESC-derived and iPSC-derived neurons cluster together, as do their parental hESC-derived and iPSC-derived NSC. These data indicate that differentiation-associated expression changes outweigh potential changes due to the reprogramming procedure (Fig. 3b). Differentially expressed genes (DEG) in iPS-Neurons were mostly associated with transcription regulation and homeobox genes. Overall, the coefficient of determination for gene expression levels in iPS-NSC and iPS-Neurons amounted to R2 = 0.75 (data not shown).
Taken together, our data indicate that isogenic hESC-derived and iPSC-derived neurons generated via a highly standardized intermediate NSC population [14] display similar DNA methylation and gene expression patterns. Thus, the reprogramming process does not result in major alterations of the methylation and gene expression patterns in iPSC-derived neurons. It is important to note though that the iPSCs in our experimental setting were generated from hESC-derived NSC, while patient-derived iPSCs are typically generated from fibroblasts or peripheral blood mononuclear cells. To assess how methylation levels observed in our system relate to the data obtained from neurons generated from fibroblast-derived iPSC, we performed a comparative in silico analysis including hESC-derived neural cells from another hESC line (H9) and fibroblast-derived neuronal populations from controls, iPD and LRRK2 PD patients [11, 15]. Hypothesis-free variant principal component analysis (PCA) of the 450 K data revealed that pluripotent stem cells, fibroblast-iPSC-derived neurons, H9 hESC-derived neural cells, and I3 hESC-derived neural cells clustered in distinct tiers (Fig. 4 and Additional file 3). PC2 was most likely associated with cell type, whereas the biggest variance (PC1) could not be attributed to a single factor. The segregation pattern of the different neuronal populations suggests that the individual genetic background (patient-derived vs. H9 ESC-derived vs. I3 ESC-derived) and differences in the neuronal differentiation protocols have a much higher impact on the methylation landscape than the reprogramming process itself. In line with this notion, all isogenic neurons generated from I3 ESC and I3 ESC-NSC-derived iPSC formed a tight cluster, which segregated from fibroblast- and H9 ESC-derived neurons (Fig. 4). We could also identify a weak, but significant correlation regarding PC1 and batch effects (p < 0.0002, R2 = 0.73). From our observations, we conclude that a thorough comparison of neurons generated from ESC-NSC- and fibroblast-derived iPSC would require isogenic human ESC and iPSC. Such comparative studies might be extended to iPSC-derived vs. central nervous system (CNS) tissue-derived neurons isolated from the same donor, which could in the end resolve to what extent the epigenetic landscape of reprogrammed neurons reflects that of primary human neurons.
DNA methylation changes during differentiation of isogenic hESC- and iPSC-derived neurons
To assess the acquisition of methylation patterns during neuronal differentiation in depth, we pooled the data of the hESC- and iPSC-derived NSC and compared them to the pooled data of hESC- and iPSC-derived neurons. We performed a paired t test (p < 0.001) applying a less conservative methylation difference of 10%, which identified 1.314 DMCG only (Fig. 5a, Additional file 4). Only 41 CpGs were hypomethylated, while the vast majority (1.273 CpGs) was hypermethylated in neurons compared to NSC (Fig. 5b). We did not identify an obvious enrichment for DMCG comparing gene sub-regions such as transcription start sites, promoter, and gene body regions in hES- and iPS-Neurons compared to NSC (Additional file 5), but less DMCG in CpG island regions compared to other annotations such as CpG shelfs or shores (Fig. 5c).
DNA methylation levels of SNCA intron 1 increased during neuronal differentiation in isogenic hESC- and iPSC-derived neurons
We next used MiSeq amplicon sequencing of bisulphite-treated genomic DNA (Bi-PROF; [16]) to validate our data obtained from the 450 K assay and to specifically assess DNA methylation at a single base level. We choose dosage-sensitive genes known to be involved in different neurological disorders, including APP (gene duplication, Alzheimer’s disease, AD), SNCA (gene duplication/triplication, Parkinson’s disease, PD) [17], and PMP22, which has been associated with Charcot-Marie-Tooth 1A (gene duplication) and several other CNS disorders [18, 19]. Epigenetic modification of such dosage-sensitive genes could represent a missing link between familial and sporadic forms of NDD [1]. Furthermore, we chose MIR886, which was found to be differentially methylated in PD [20] and GNAS, a gene with a highly complex imprinted expression pattern [21] (Additional file 6). All aforementioned samples (hES-Neuron and iPS-Neuron clones 1–3) were included in the analysis. Pearson correlation coefficients demonstrated a correlation of 0.92 for individual CpGs detected by 450 K bead array and Bi-PROF, respectively. After a successful validation of single CpGs obtained from the 450 K assay, we analyzed amplicons located at the promoter regions and CpG islands of the candidate genes (Additional file 6). A comparison of hESC- and iPSC-derived neurons revealed that most regulatory regions exhibited unchanged DNA methylation patterns; the DNA methylation levels of APP, GNAS, MIR886, and SNCA promoter and intron 1 were similar comparing hESC- and iPSC-derived neurons (Fig. 6, Additional file 7). Average methylation levels over all CpGs were also comparable between hESC- and iPSC-derived NSC (Fig. 6, Additional file 7). The pattern within the SNCA intron 1 displayed higher mean methylation levels over all analyzed CpGs compared to the canonical promoter region in line with the previous analysis of bulk native brain tissue (Additional file 7) [2, 5]. These data show that DNA methylation levels at promoter regions of the selected candidate genes are similar in ESC- and iPSC-derived neurons.
DNA methylation alterations during neuronal differentiation of NSCs were observed specifically at the SNCA intron 1, but not for the other selected and analyzed NDD-associated genes. HES-NSC generated from I3 hESC and iPS-NSC generated from I3 hES-NSC-derived iPSC exhibited slightly higher, though non-significant (iPSC vs. iPS-NSC p = 0.08) DNA methylation levels of SNCA intron 1 than their parental PSC populations and hESC of the line I6; differentiation of the iPSC-derived NSC line I3 resulted in a further significant increase in DNA methylation (ESC vs. iPS-Neurons p = 0.03, iPSC vs. iPS-Neurons p = 0.02, iPS-NSC vs. iPS-Neurons p = 0.02; Additional file 8). To obtain more insight into DNA methylation changes in the SNCA intron 1 across neuronal differentiation, we analyzed NSC and neurons differentiated for 2, 4, and 6 weeks (Fig. 7a). Mean DNA methylation levels over all analyzed CpG sites of SNCA intron 1 increased during a 6-week period of neuronal differentiation in both hESC-derived (2.3-fold) and iPSC-derived neurons (3.5-fold, p = 0.016; Fig. 7a). While DNA methylation levels increased at each CpG upon neuronal differentiation, CpGs 3, 5, and 20 showed prominent changes (p = 0.001, p = 0.002, and p = 0.008, respectively; Fig. 7a). Remarkably, even at this single CpG level, methylation changes in iPSC- and hESC-derived neurons were very similar and showed a comparable pattern (Fig. 7a). Interestingly, increasing DNA methylation of SNCA intron 1 across neuronal differentiation was not associated with decreased but increased SNCA expression in both hESC- and iPSC-derived neurons (Fig. 7b).