Genome-wide DNA methylation studies are becoming increasingly used in search of etiological factors contributing to complex non-Mendelian disease, as the susceptibility of DNA methylation to environmental influences and its potential for metastable drift may account for complex disease features, such as a discordance of monozygotic twins, parent of origin effects, an unequal frequency of affected males and females, complex inheritance, and a late age at onset, among others [1
]. DNA methylation changes in the brain are becoming increasingly recognized as important mediators of behavioral phenotypes in model organisms and psychiatric disease in humans [4
Despite the likelihood of epigenetic changes as etiological factors contributing to psychiatric disease risk, the success of the first round of epigenomic studies has been limited [8
]. In the first epigenomic profiling studies performed in major psychosis, Mill et al. found moderate fold changes in prefrontal cortex DNA methylation. In the WDR18
glutamate receptor subunit gene, an 8% DNA methylation difference was detected between males with schizophrenia and controls, while female patients with bipolar disorder were 6% more methylated than controls at the RPL39
]. No significant differences were found in an analysis of 50 loci in temporal cortex of schizophrenia affected individuals [10
]. A recent methylome profiling study in major depressive disorder (MDD) did not identify any significant loci after correction for multiple testing; however, they did successfully validated a 60% of the top nominally significant differences [11
]. Of these, the largest depression associated effect size was 22%.
A consistent feature of these studies is the low effect size associations detected in the brain. A probable explanation for these observations is that true disease differences exist in a subpopulation of cells that are subject to dilution by disease non-relevant cell types, a factor particularly relevant in the brain, which represents one of the most cellularly heterogeneous organs in the body. This situation calls for algorithms capable of detecting DMRs of small effect size in order to direct downstream validation and follow up functional studies, such as cell type specific analyses. In this regard, the ability of a DMR detection technique to adjust for covariates such as cellular heterogeneity, medication status, or age are of particular interest in psychiatric phenotypes but to date, few available algorithms for DMR detection allow for these adjustments.
Another factor that remains at issue is that phenotypically relevant epigenetic changes may occur over relatively small regions. A number of locus specific studies highlight the importance of short genomic regions in regulating phenotypic outcome. Epigenetic changes spanning short genomic regions have been identified in imprinting control regions, over exonic regions that may direct alternative splicing, and at transcription factor binding sites that have been associated with early life trauma exposure or major psychosis [9
]. The power to identify short DMRs is an important facet of DMR finding algorithms used in studies searching for small epigenetic aberrations conferring phenotypic variation.
The application of tiling array technology to the study of DNA methylation has greatly increased the resolution over earlier microarray based technologies and added to the potential to discover novel epigenetic changes. Tiling array experiments are based on measuring the genomic locations of enriched DNA fragments that hybridize across adjacently located probes called tiles. The experiments performed prior to hybridization involve enriching for the molecular marker of interest, either through antibody based immunoprecipitation employed in ChIP-chip [15
], MeDIP [16
], or through enzymatically selecting a portion of the genome, such as with methylation sensitive restriction enzymes as is employed in numerous DNA methylome techniques [18
]. The enriched fractions are fragmented to improve target specificity, generally to lengths of 50–200 base pairs. After microarray hybridization, the combinatorial effects of fragment binding to specific genomic locations will result in peaks of signal intensity after data processing that may be detected by downstream data analysis applications.
A number of excellent programs that contain peak finding algorithms are available for the analysis of tiling array data, some of which include Ringo [22
], ChiPOTle [23
], CHARM [19
], TileMap [24
], ACME [25
], and MPEAK [26
], among others. There is a large degree of variation in the statistical methods employed, the ease of use, and the versatility across multiple experiment types. For example, many of these algorithms, such as CHARM and Ringo, were designed for one type of platform, such as NimbleGen arrays, but can now be applied to other datasets. Others, such as ChiPOTle are limited in the number of probes that can be analyzed (IE: 60,000), which makes it difficult to apply to larger tiling array datasets. With the exception of CHARM, these DMR finding algorithms are confined to the investigation of group classifiers as opposed to quantitative variables such as multiple treatment doses or age and do not allow for the correction of covariates prior to peak identification and statistical evaluation.
Cumulatively, the application of tiling array analyses to DNA methylation in heterogeneous tissues, such as brain, require the ability to detect DMRs of small effect size and of short length. A simple analysis paradigm applicable to multiple microarray platforms and satisfying these requirements will add to the successful identification of phenotypically relevant epigenetic variation across a diverse range of phenotypes. To address these issues we present an open source, freely available Perl application referred to as “Binding Intensity Only Tiling array analysis” or “BioTile”. The BioTile algorithm is ideally suited to the identification of small length and low effect size DMRs, while not sacrificing power to detect longer DMRs, and is applicable across a range of tiling microarray platforms.