Conserved and divergent features of DNA methylation in embryonic stem cell-derived neurons

DNA methylation functions in genome regulation and is implicated in neuronal maturation. Early post-natal accumulation of atypical non-CG methylation (mCH) occurs in neurons of mice and humans, but its precise function remains unknown. Here we investigate mCH deposition in neurons derived from mouse ES-cells in vitro and in cultured primary mouse neurons. We find that both acquire comparable levels of mCH over a similar period as in vivo. In vitro mCH deposition occurs concurrently with a transient increase in Dnmt3a expression, is preceded by expression of the post-mitotic neuronal marker Rbfox3 (NeuN) and is enriched at the nuclear lamina. Despite these similarities, whole genome bisulfite sequencing reveals that mCH patterning in mESC-derived neurons partially differs from in vivo. mESC-derived neurons therefore represent a valuable model system for analyzing the mechanisms and functional consequences of correct and aberrantly deposited CG and non-CG methylation in neuronal maturation.


Introduction 40
The unique epigenomic landscape of neurons is hypothesized to allow these post- In vivo, mouse cortical neurons begin to acquire readily detectable levels of mCH around 2 100 weeks after birth, which continues to increase up to 6 weeks of age and remains high 101 throughout adulthood (Lister et al., 2013). To assess whether this can be recapitulated in vitro, 102 we used two independent approaches. First, we isolated primary cortical and hippocampal 103 neurons from day 18 embryonic C57BL/6 mice (E18, average gestation 18.5 days) and 104 cultured these for up to 14 days in vitro (DIV). We hypothesized that if mCH accumulation was 105 due to cell intrinsic developmentally hardwired processes, 14DIV should correlate with the 106 temporal acquisition of mCH in vivo ( Figure 1A). In the second approach, we adapted an 107 established differentiation protocol to generate mouse cortical neurons from embryonic stem 108 cells (mESCs) (Bibel et al., 2007). We hypothesized that if this developmental model 109 recapitulated neural development and neuronal maturation in vivo, mCH would occur within 110 several weeks ( Figure 1B). Two different mouse ESC lines, R1 (Nagy et al., 1993)  in either low K + or high K + buffer. Low levels of tracer endocytosis into synaptic vesicles in low 162 K + was superseded by high levels of bulk endocytosis in depolarised cells, suggesting a strong 163 rapid burst of neuroexocytosis and compensatory endocytosis (Cousin, 2009  including glial cells (see Figure 1K), as described previously (Bibel et al., 2004). 208 Immunocytochemical (ICC)-based analysis of DNA methylation in NeuN-positive cells 209 revealed an increase in the level of nuclear mCA labeling between days 18 and 28, which 210 remained high to day 38 ( Figure 1G, H). As the initial observation of mCA was significantly 211 later than the initial observation of NeuN, a post-mitotic phenotype is likely a prerequisite for 212 subsequent acquisition of mCA. Consistent with this, we found only minimal labelling for mCA 213 in Pax6-positive neural progenitors differentiated for 9-10 days relative to ~2-fold higher levels 214 in NeuN-positive neurons differentiated for 28-38 days ( Figure 1I, J). Similarly, we found 215 minimal labelling for mCA in GFAP-positive glial cells and in additional unidentified cell types 216 within the cultures that did not label for either neuronal or glial markers ( Figure 1K, L). 217 Methylation of CH sites has been shown to be catalysed by Dnmt3a in vivo (Feng et 218 al., 2005, Stroud et al., 2017. We therefore analysed the transcript abundance of Rbfox3 219 (NeuN) and the DNA methyltransferases Dnmt3a and Dnmt1 during differentiation by RT-220 qPCR ( Figure 1M). Consistent with the results of the ICC, Rbfox3 (NeuN) expression was 221 found to significantly increase between days 9 and 18, reaching a plateau around day 28. In 222 contrast, and in agreement with previous data from both primary neurons and mouse brain 223 During our ICC analysis of mCA labelling, we observed that this DNA modification was 235 enriched near the nuclear periphery in our in vitro neuronal cultures. To determine whether 236 this localisation was unique to mCA, ESC-derived neurons were co-labelled for total 5-237 methylcytosine (5mC) using an antibody predicted to label both mCG and mCH, and with the 238 mCA-specific antibody, and the intranuclear distribution of the two compared ( Figure 2A). 239 Consistent with a more restricted intranuclear distribution of mCA, total 5mC labelling was 250 observed to be more broadly distributed within the nucleus. While both marks showed a diffuse 251 labeling, mCA was highly enriched at the nuclear periphery, whereas 5mC additionally strongly 252 labelled intranuclear foci, which were devoid of mCA labelling. The intensely labelled 5mC foci 253 were found to also stain strongly with DAPI ( Figure 2A To directly examine the distribution of DNA methylated in specific sequence contexts, 257 neurons were labelled for either mCA or using an antibody specific for mCG ( Figure 2B). In 258 addition, we examined localisation of 5-hydroxymethylcytosine (5hmC) ( Figure 2B Next, we assessed regional correlation in mCG and mCH levels in 100 kb bins of the 528 whole genome (excluding chromosomes X and Y) between mESC-derived neurons, fetal 529 frontal cortex, and 7-week old mouse PFC neurons and glia ( Figure 5D and 6D). For mCG, 530 fetal brain and adult neurons were the most similar, with adult glia joining at the next node, 531 while mESC-derived neurons formed their own branch. The low correlation between neuronal 532 samples is likely due to the overall higher methylation of CG in ESC-derived neurons. For 533 mCH, there was a high similarity between in vitro-and in vivo-derived neuronal datasets, while 534 glia were similar to the fetal sample, consistent with the neuron-specific accumulation of mCH. 535 This clustering based on bins also showed that differences in methylation between neurons 536 were not evenly distributed throughout the genome but show regional variability. 537 538 539

Supplementary Figure 7 DNA methylation in CG context in gene bodies sorted for 540 differences in mCH between neuronal samples 541
Genes are in the same order based on CH methylation difference as in Fig 5B but showing 542

CG methylation for gene bodies and flanking 10 kb for fetal mouse frontal cortex (fetal), NeuN-543 negative cells from 7-week adult mouse prefrontal cortex (glia), 7-week adult mouse prefrontal 544 cortex neurons (in vivo neurons), and d38 mESC-derived neurons (in vitro neurons). 545
Difference in mCH between both neuronal samples used for gene order is shown on the right. vivo neuron populations and discriminate them against fetal or glial cells, we applied GSEA 576 on genes ranked by a combination of similarity between the mESC-derived in vitro neurons 577 and in vivo neurons, and dissimilarity to non-neuronal cell types (glia and fetal brain cells, 578 Figure 7). This analysis was performed for mCG and mCH independently and resulted in an 579 enrichment for pathways linked to genes that have an equivalent methylation 580 was continued for 8 days, medium was changed every two days, and supplemented with 5 803 µM retinoic acid for the final 4 days. Cell aggregates were dissociated using Accutase, single 804 cell selected through a 40 µm cell strainer, and plated onto 0.1 mg/ml poly-ornithine / 4 µg/ml 805 laminin-coated plates in N2 medium (DMEM/F12, 1x Glutamax 1x N2 supplement, 20 µg/ml 806 insulin, 50 µg/ml BSA) at a density of 50,000-100, 000 cells/cm 2 . After 2 days, cells were 807 changed into N2B27 (Neurobasal, 1x N2 supplement, 1x B27 supplement, 1x Glutamax). 808 N2B27 was changed every 4 days for 12 days, following which 50% media changes were 809 performed for the remaining culture time. 810

Supplementary Figure 9 Top 50 enriched pathways for genes differentially methylated 582 in CG context between in vitro neurons and in vivo neurons
High potassium-mediated depolarisation was performed as previously described 811

RT-qPCR 845
Total RNA was extracted using the Macherey-Nagel NucleoSpin RNA kit, including on 846 column digestion of DNA with RNase-free ase according to anufacturer's specifications. 847 Concentration and 260/280 ratios were quantified using a NanoDrop 1000 spectrophotometer 848 before cDNA synthesis using the iScript cDNA synthesis kit (Bio-Rad). Primers were designed 849 to span exon-exon boundaries wherever possible (Supplementary Table 2). When this was 850 not possible, samples were excluded if genomic DNA contamination was more than 10-fold 851 over the cDNA concentration. Quantitative PCR (qPCR) reactions used SsoFast Evagreen 852 (Bio-Rad) with c A te plate according to anufacturer's instructions, using a C1000 853 Thermocycler (Bio-Rad) and CFX software. Results were analysed as described previously 854 (Livak and Schmittgen, 2001). Phase contrast microscopy was performed on live cells using an Olympus IX51. 891 Confocal microscopy was performed on fixed cells using either a Zeiss 710 confocal 892 microscope and a 40x water immersion objective or a Leica SP8 confocal microscope using 893 a 60x oil immersion objective. High throughput imaging was performed using a Perkin-Elmer 894 Operetta equipped with a 20x Air objective. Brain section imaging was performed with Nikon 895 A1-R confocal microscope and a 60x 1.4 NA oil immersion objective. 896 Image analysis was performed by manual masking of nuclei and measuring 897 fluorescence intensity/nucleus using Image J ( Figure 1J The Gene Ontology, 2019). Pathways were sorted by NES score (enrichment score 950 normalized to mean enrichment of random samples of the same size) and only pathways with 951 p-value < 0.05 were considered in subsequent analyses. 952 For similarity analysis, every gene was given a score as follows: First, the difference 953 in weighted DNA methylation between in vivo and in vitro neurons (y) was calculated for each 954 gene and scaled to a value between 0 and 1 (using the formula x = |y-1|), so that genes that 955 were more similar in methylation state between both samples would have a value (x) closer to 956 1. Then, the average of weighted DNA methylation per gene in neuronal samples was 957 compared against the average weighted DNA methylation of glial and fetal samples, resulting 958 again in a score between 0 and 1, with genes showing greater differences having a value 959 closer to 1. Both scores for similarity between neurons and dissimilarity to non-neuronal 960 samples were added in order to give both aspects the same weight, and scaled to values 961 between 0 and 1, with values closer to 1 representing genes being more similar in methylation 962 state between both neuronal samples but different compared to glia and fetal frontal cortex. 963 This similarity score was then used to rank all genes for GSEA using the fgsea package as 964 described above. 965 966 Fluorescence-activated nuclear sorting 967 Intact nuclei were isolated from cell pellets as described previously (Li et al., 2014, 968 Okada et al., 2011). Briefly, cells were Dounce-homogenised on ice in chilled nuclear 969