Limits...
Identification and systematic annotation of tissue-specific differentially methylated regions using the Illumina 450k array.

Slieker RC, Bos SD, Goeman JJ, Bovée JV, Talens RP, van der Breggen R, Suchiman HE, Lameijer EW, Putter H, van den Akker EB, Zhang Y, Jukema JW, Slagboom PE, Meulenbelt I, Heijmans BT - Epigenetics Chromatin (2013)

Bottom Line: Various studies have compared tissues to characterize epigenetically regulated genomic regions, but due to differences in study design and focus there still is no consensus as to the annotation of genomic regions predominantly involved in tissue-specific methylation.Further analysis revealed that these regions were associated with alternative transcription events (alternative first exons, mutually exclusive exons and cassette exons).We conclude that tDMRs preferentially occur in CpG-poor regions and are associated with alternative transcription.

View Article: PubMed Central - HTML - PubMed

Affiliation: Molecular Epidemiology, Leiden University Medical Center, Leiden, The Netherlands. B.T.Heijmans@lumc.nl.

ABSTRACT

Background: DNA methylation has been recognized as a key mechanism in cell differentiation. Various studies have compared tissues to characterize epigenetically regulated genomic regions, but due to differences in study design and focus there still is no consensus as to the annotation of genomic regions predominantly involved in tissue-specific methylation. We used a new algorithm to identify and annotate tissue-specific differentially methylated regions (tDMRs) from Illumina 450k chip data for four peripheral tissues (blood, saliva, buccal swabs and hair follicles) and six internal tissues (liver, muscle, pancreas, subcutaneous fat, omentum and spleen with matched blood samples).

Results: The majority of tDMRs, in both relative and absolute terms, occurred in CpG-poor regions. Further analysis revealed that these regions were associated with alternative transcription events (alternative first exons, mutually exclusive exons and cassette exons). Only a minority of tDMRs mapped to gene-body CpG islands (13%) or CpG islands shores (25%) suggesting a less prominent role for these regions than indicated previously. Implementation of ENCODE annotations showed enrichment of tDMRs in DNase hypersensitive sites and transcription factor binding sites. Despite the predominance of tissue differences, inter-individual differences in DNA methylation in internal tissues were correlated with those for blood for a subset of CpG sites in a locus- and tissue-specific manner.

Conclusions: We conclude that tDMRs preferentially occur in CpG-poor regions and are associated with alternative transcription. Furthermore, our data suggest the utility of creating an atlas cataloguing variably methylated regions in internal tissues that correlate to DNA methylation measured in easy accessible peripheral tissues.

No MeSH data available.


Related in: MedlinePlus

Example of the tDMR finder algorithm used for the HOXD3 gene. Tissue-specific differentially methylated regions were identified in a two-step approach: first, we identified tDMPs. CpGs were considered to be tDMPs when there was a genome-wide significant mean difference of ≥ 10%. The mean difference was expressed as a mean sum of squares. A difference ≥ 10% equals a mean sum of squares ≥ 0.01 (square of 10% = 0.12). To test whether the difference was significant, we applied a linear model per CpG site, with a random effect for each individual to correct for any inter-individual variation. From this linear model we obtained a P value (F-test) per CpG site and used a multiple testing corrected P value as a cut-off (10-7). Second, we identified tDMRs as regions with at least three tDMPs with an inter-CpG distance of at most 1 kb and a maximum of three non-tDMPs. Mb, megabase; tDMP, tissue-specific differentially methylated position; tDMR, tissue-specific differentially methylated region.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3750594&req=5

Figure 1: Example of the tDMR finder algorithm used for the HOXD3 gene. Tissue-specific differentially methylated regions were identified in a two-step approach: first, we identified tDMPs. CpGs were considered to be tDMPs when there was a genome-wide significant mean difference of ≥ 10%. The mean difference was expressed as a mean sum of squares. A difference ≥ 10% equals a mean sum of squares ≥ 0.01 (square of 10% = 0.12). To test whether the difference was significant, we applied a linear model per CpG site, with a random effect for each individual to correct for any inter-individual variation. From this linear model we obtained a P value (F-test) per CpG site and used a multiple testing corrected P value as a cut-off (10-7). Second, we identified tDMRs as regions with at least three tDMPs with an inter-CpG distance of at most 1 kb and a maximum of three non-tDMPs. Mb, megabase; tDMP, tissue-specific differentially methylated position; tDMR, tissue-specific differentially methylated region.

Mentions: Tissue types tended to cluster together according to genome-wide DNA methylation data indicating the occurrence of tissue-specific methylation patterns (Additional file 2: Figure S1E, F). To study these patterns in more detail, we developed an algorithm to identify tissue-specific differentially methylated regions systematically using 450k methylation data as described in Figure 1 (also see Methods). Briefly, first tissue-specific differentially methylated positions (tDMPs) were identified. tDMPs were defined as CpGs with a DNA methylation difference between tissues that was: (1) genome-wide significant (P < 10-7) and (2) had a mean sum of squares ≥ 0.01 (equals (10%)2, that is, the mean of the difference between the individual tissues and the overall mean across tissues should be greater than 10%). Next, differentially methylated regions (DMRs) were identified as regions with at least three differentially methylated positions (DMPs) with an inter-CpG distance ≤ 1 kb, interrupted by at most three non-DMPs across the whole DMR (see Methods; the algorithm is in Additional file 4). The algorithm detected 3,533 and 5,382 tDMRs in the peripheral and internal tissue datasets, respectively (Table 1 and Additional file 5: Table S2). There were 4,877 unique (that is, non-overlapping) tDMRs between datasets. Interestingly, 2,019 tDMRs were detected in both peripheral and internal tissues (9,388 CpGs in common, P < 0.001). The tDMR distribution over the genome was similar for the two datasets (Additional file 3: Figure S2C). A further indication of the validity of the tDMRs was obtained from a visualization of the tDMRs in a heat map according to tissue, which showed the expected clustering by germ layer and confirmed the previously reported cellular similarities between blood and saliva, and between hair and buccal swabs (Additional file 6: Figure S3) [16].


Identification and systematic annotation of tissue-specific differentially methylated regions using the Illumina 450k array.

Slieker RC, Bos SD, Goeman JJ, Bovée JV, Talens RP, van der Breggen R, Suchiman HE, Lameijer EW, Putter H, van den Akker EB, Zhang Y, Jukema JW, Slagboom PE, Meulenbelt I, Heijmans BT - Epigenetics Chromatin (2013)

Example of the tDMR finder algorithm used for the HOXD3 gene. Tissue-specific differentially methylated regions were identified in a two-step approach: first, we identified tDMPs. CpGs were considered to be tDMPs when there was a genome-wide significant mean difference of ≥ 10%. The mean difference was expressed as a mean sum of squares. A difference ≥ 10% equals a mean sum of squares ≥ 0.01 (square of 10% = 0.12). To test whether the difference was significant, we applied a linear model per CpG site, with a random effect for each individual to correct for any inter-individual variation. From this linear model we obtained a P value (F-test) per CpG site and used a multiple testing corrected P value as a cut-off (10-7). Second, we identified tDMRs as regions with at least three tDMPs with an inter-CpG distance of at most 1 kb and a maximum of three non-tDMPs. Mb, megabase; tDMP, tissue-specific differentially methylated position; tDMR, tissue-specific differentially methylated region.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3750594&req=5

Figure 1: Example of the tDMR finder algorithm used for the HOXD3 gene. Tissue-specific differentially methylated regions were identified in a two-step approach: first, we identified tDMPs. CpGs were considered to be tDMPs when there was a genome-wide significant mean difference of ≥ 10%. The mean difference was expressed as a mean sum of squares. A difference ≥ 10% equals a mean sum of squares ≥ 0.01 (square of 10% = 0.12). To test whether the difference was significant, we applied a linear model per CpG site, with a random effect for each individual to correct for any inter-individual variation. From this linear model we obtained a P value (F-test) per CpG site and used a multiple testing corrected P value as a cut-off (10-7). Second, we identified tDMRs as regions with at least three tDMPs with an inter-CpG distance of at most 1 kb and a maximum of three non-tDMPs. Mb, megabase; tDMP, tissue-specific differentially methylated position; tDMR, tissue-specific differentially methylated region.
Mentions: Tissue types tended to cluster together according to genome-wide DNA methylation data indicating the occurrence of tissue-specific methylation patterns (Additional file 2: Figure S1E, F). To study these patterns in more detail, we developed an algorithm to identify tissue-specific differentially methylated regions systematically using 450k methylation data as described in Figure 1 (also see Methods). Briefly, first tissue-specific differentially methylated positions (tDMPs) were identified. tDMPs were defined as CpGs with a DNA methylation difference between tissues that was: (1) genome-wide significant (P < 10-7) and (2) had a mean sum of squares ≥ 0.01 (equals (10%)2, that is, the mean of the difference between the individual tissues and the overall mean across tissues should be greater than 10%). Next, differentially methylated regions (DMRs) were identified as regions with at least three differentially methylated positions (DMPs) with an inter-CpG distance ≤ 1 kb, interrupted by at most three non-DMPs across the whole DMR (see Methods; the algorithm is in Additional file 4). The algorithm detected 3,533 and 5,382 tDMRs in the peripheral and internal tissue datasets, respectively (Table 1 and Additional file 5: Table S2). There were 4,877 unique (that is, non-overlapping) tDMRs between datasets. Interestingly, 2,019 tDMRs were detected in both peripheral and internal tissues (9,388 CpGs in common, P < 0.001). The tDMR distribution over the genome was similar for the two datasets (Additional file 3: Figure S2C). A further indication of the validity of the tDMRs was obtained from a visualization of the tDMRs in a heat map according to tissue, which showed the expected clustering by germ layer and confirmed the previously reported cellular similarities between blood and saliva, and between hair and buccal swabs (Additional file 6: Figure S3) [16].

Bottom Line: Various studies have compared tissues to characterize epigenetically regulated genomic regions, but due to differences in study design and focus there still is no consensus as to the annotation of genomic regions predominantly involved in tissue-specific methylation.Further analysis revealed that these regions were associated with alternative transcription events (alternative first exons, mutually exclusive exons and cassette exons).We conclude that tDMRs preferentially occur in CpG-poor regions and are associated with alternative transcription.

View Article: PubMed Central - HTML - PubMed

Affiliation: Molecular Epidemiology, Leiden University Medical Center, Leiden, The Netherlands. B.T.Heijmans@lumc.nl.

ABSTRACT

Background: DNA methylation has been recognized as a key mechanism in cell differentiation. Various studies have compared tissues to characterize epigenetically regulated genomic regions, but due to differences in study design and focus there still is no consensus as to the annotation of genomic regions predominantly involved in tissue-specific methylation. We used a new algorithm to identify and annotate tissue-specific differentially methylated regions (tDMRs) from Illumina 450k chip data for four peripheral tissues (blood, saliva, buccal swabs and hair follicles) and six internal tissues (liver, muscle, pancreas, subcutaneous fat, omentum and spleen with matched blood samples).

Results: The majority of tDMRs, in both relative and absolute terms, occurred in CpG-poor regions. Further analysis revealed that these regions were associated with alternative transcription events (alternative first exons, mutually exclusive exons and cassette exons). Only a minority of tDMRs mapped to gene-body CpG islands (13%) or CpG islands shores (25%) suggesting a less prominent role for these regions than indicated previously. Implementation of ENCODE annotations showed enrichment of tDMRs in DNase hypersensitive sites and transcription factor binding sites. Despite the predominance of tissue differences, inter-individual differences in DNA methylation in internal tissues were correlated with those for blood for a subset of CpG sites in a locus- and tissue-specific manner.

Conclusions: We conclude that tDMRs preferentially occur in CpG-poor regions and are associated with alternative transcription. Furthermore, our data suggest the utility of creating an atlas cataloguing variably methylated regions in internal tissues that correlate to DNA methylation measured in easy accessible peripheral tissues.

No MeSH data available.


Related in: MedlinePlus