Allele-specific bisulfite sequencing of DMRs

To address if parental alleles maintained their own methylation pattern in F1-hybird animals I performed traditional bisulfite sequencing, since the MS analysis did not discriminate between the two different parental alleles. I chose DMR regions where a characteristic sequence difference remained after bisulfite-treatment of genomic DNA and analyzed the tissue where the greatest methylation difference between parental strains was observed. In total, six regions were analyzed; four containing single nucleotide polymorphisms (Coro2a, Isoc2b, 1600021P15Rik and Asb4) and two regions harboring small insertions in parental C57BL/6 (Slc27a6) or BALB/c (Pdgfrb) sequences. Besides two male F1-hybrid animals (one from each mating combination) I also included two male individuals from each parental strain. Genomic DNA, derived from BMMs (or testis in case of Pdgfrb) was treated with sodium-bisulfite, similar to the MassARRAY experiments (see methods section 4.5.3, page 57). PCR products of these five regions were generated using the same primer pairs, specific to bisulfite-treated DNA as well as the same PCR conditions which have been used in the MassARRAY experiments. PCR-products were then cloned into the pCR®2.1TOPO-vector and transfected in E. coli. Insert-containing plasmids, derived from single colonies were further purified and sequenced. Sequencing results for five amplicons are shown in Figure 5.25. The figure illustrates that in the parental strains the genes Coro2a, Isoc2b, Asb4, 1600021P15Rik and Slc27a6 were methylated in BALB/c, whereas Pdgfrb was methylated in C57BL/6. This observation was in line with the microarray and MassARRAY results. In F1-hybrid animals the methylation pattern within one animal was mixed, with some sequences being methylated and some unmethylated. Both F1-mice used showed the same methylation distribution pattern between individually cloned sequences and were, therefore, combined to one pool.

Furthermore, since I restricted this analysis to amplicons, which can be traced back to their strain origin, I was able to determine the strain parentage of a specific amplicon within the F1-mice. Hence, alleles of Coro2a, Pdgfrb, Asb4, 1600021P15Rik and Slc27a6 that derived from C57BL/6 exhibited a very similar methylation pattern as seen in the parental strain. The same was observed for alleles, which can be traced back to the BALB/c origin. However, the Isoc2b gene uniquely showed a different pattern, where both alleles within F1-hybrid animals contained unmethylated as well as methylated sequences. This pattern was obvious in both F1-mice examined in this experiment.

Figure 5.25 Strain-specific methylation patterns are mainly controlled in cis. Allele-specific bisulfite sequencing of DMRs in Coro2a (A, controlled in cis), Pdgfrb (B, controlled in cis), Isoc2b (C, controlled in trans), 1600021P15Rik (D, controlled in cis), Asb4 (E, controlled in cis) and Slc27a6 (F, controlled in cis) (D-F are shown on the next page). The genomic position of CpGs within the amplicons is shown at the top. Sequence variations used to distinguish the different parental alleles are marked in blue for C57BL/6 and in red for BALB/c. Individual CpGs are represented by either white (unmethylated) or black (methylated) squares. Lines of squares represent independently sequenced clones derived from two independent sample preparations, derived from reciprocal crosses.

Taken together, in 5 (out of 6) regions the allelic methylation patterns in F1 hybrids were established essentially as observed on the parental alleles, suggesting that DNA-methylation at these sites is largely controlled in cis by the local genetic sequence.

Amongst the studied regions, only at the Isoc2b promoter, a mixed DNA methylation pattern was observed with some sequences being unmethylated and others methylated independent of their origin. Therefore the differential pattern must be controlled in trans, for example through a yet unknown strain-specific epigenetic modifier.

Figure 5.25 continued from previous page.

6 6 D D i i s s c c u u s s s s i i o o n n

Epigenetic phenomena in general and DNA methylation in particular play a major role in the regulation of gene expression. The latter mechanism contributes to both the shaping of cellular phenotypes within an organism and the establishing of interindividual differences between organisms. In addition, differences in DNA methylation can not only be observed in transformed cells or other non-cancer diseases, but also at very different levels in various tissue types and developmental stages. In this work, I have established methods to study the contribution of differential DNA methylation to cell type-specific gene expression as well as the occurrence of differentially methylated regions between individuals (inbred mice strains).

6.1 Mapping DNA methylation

In the past years epigenetic modifications such as DNA methylation have been intensively studied. Several methodologies have been developed to map DNA methylation, each of which having its own advantages and disadvantages as highlighted in the introduction. As part of this thesis, methyl-CpG immunoprecipitation (MCIp)¹¹⁰ was modified and utilized to map DNA methylation both on single-gene level and on a global scale by combination with quantitative real-time PCR or microarrays, respectively. MCIp is based on the recombinant antibody-like MBD2–Fc fusion protein that was originally designed by our lab to globally detect disease-related hypermethylation in CpG islands on a global level¹¹⁰. A crucial step for accurate DNA methylation mapping is the sample pretreatment. Sanger sequencing of bisulfite treated genomic DNA, which is the standard method, provides the best resolution, because it maps methylated cytosines at single base-pair resolution.

Large-scale bisulfite-DNA sequencing has been successfully initiated^153,154, but is very laborious and resource-intensive. Alternative, less direct methods have been developed to map DNA methylation on a global scale. Approaches based on methylation-sensitive restriction enzymes enrich fragment due to digestion of methylated^118,119 or unmethylated

116,155

using MBD affinity purification (MAP, using the MBD domain for binding of methylated DNA)^107,123. Both methods gave comparable results¹²⁰, but require a relatively high density of DNA methylation such as methylated CGIs⁷. A method to enrich unmethylated DNA using CxxC affinity purification (CAP, with x being any residue) was also recently published¹²⁶ but is also insensitive to regions with low CpG content⁷. In contrast, MCIp can divide the bulk of genomic DNA fragments into separate fractions of increasing methylation density. Therefore, during the MCIp procedure, not only the methylated DNA can be enriched, but also unmethylated DNA is recovered without sample loss. This allows the simultaneous analysis of the whole range of DNA methylation density, including both hyper- and hypomethylated DNA.

Samples enriched by either of these techniques can be globally analyzed using DNA microarrays or by direct large-scale sequencing techniques. The latter, such as 454 sequencing (Roche) or Solexa bisulfite sequencing (Illumina), are relatively fast and cheap and have been used in a number of publications^156-158. However, these readout techniques are still resource intensive and analysis cannot be restricted to distinct regions.

Microarrays are cheaper and more flexible, since one can either use custom-designed or pre-designed commercially available tiling arrays, depending on the genomic regions of interest. However, compared to large-scale sequencing approaches, microarrays receive data with only moderate resolution.

During this thesis, the MCIp-on-chip application (MCIp combined with microarray readout) originally designed by our lab to globally detect disease-related hypermethylation in CpG islands¹¹⁰ was further developed to a novel application. This work was designed as a pilot study to demonstrate the applicability of this method for mapping of cell type-specific hypomethylation. To overcome resolution restrictions and to validate the microarray analysis, the methylation levels were additionally quantified for some regions using single-gene MCIp and the standard bisulfite sequencing approach as a MCIp-independent approach. In summary, our data showed a good correlation between tissue-specific gene expression and promoter hypomethylation. Hence, the MCIp-on-chip was successfully applied to a novel application and seems to be a reliable method to detect cell type-specific hypomethylation.

Furthermore, by using MCIp I successfully fractionated genomic DNA into methylated and unmethylated pools, which has recently been performed by our group in another study as well¹⁵⁹. In their study, Schmidl and colleagues were able to identify and functionally characterize differentially methylated regions between two different human T-cell populations. In the present study the performance of this “mirror image” approach was improved by using more stringent hybridization conditions to identify differentially methylated regions (DMRs) in specific, selected genomic intervals. Since the "mirror

image" data correlated well with the MassARRAY results, this elaborate approach seems to be a reliable method for locus-wide comparative methylation analysis. With the pre-selection of differentially expressed genes and the downstream high-throughput analysis of selected CpGs by mass spectrometry, this work systematically identified DMRs and yielded new insights in individual (allele-specific) DNA methylation. Moreover, the vCGH analysis allowed the identification of genomic alternations, such as copy number variations (CNV) or deletions. Some of the larger CNV‟s, which have been validated in this thesis by qRT-PCR, were previously identified using standard comparative genomic hybridization approaches (CGH)¹⁵⁰. Standard CGH approaches utilize genomic DNA without pre-treatment steps (like MCIp), to screen for genetic changes (gains or losses) on a regular basis^160,161. The fact that these copy number variations between the two mice strains were correctly identified strongly suggests that the MCIp technique fractionated genomic DNA without sample loss like in MAP or MeDIP.

Taken together, the MCIp technique seems to be a reliable methodology to map differential DNA methylation at the level of both hypo- and hypermethylation.