Successful computational prediction of novel imprinted genes from epigenomic features

Abstract

Approximately 100 mouse genes undergo genomic imprinting, whereby one of the two parental alleles is epigenetically silenced. Imprinted genes influence processes including development, X chromosome inactivation, obesity, schizophrenia, and diabetes, motivating the identification of all imprinted loci. Local sequence features have been used to predict candidate imprinted genes, but rigorous testing using reciprocal crosses validated only three, one of which resided in previously identified imprinting clusters. Here we show that specific epigenetic features in mouse cells correlate with imprinting status in mice, and we identify hundreds of additional genes predicted to be imprinted in the mouse. We used a multitiered approach to validate imprinted expression, including use of a custom single nucleotide polymorphism array and traditional molecular methods. Of 65 candidates subjected to molecular assays for allele-specific expression, we found 10 novel imprinted genes that were maternally expressed in the placenta.

FIG. 1.

Correlation of features in each domain with imprinting. For the 11 domains examined, the correlation coefficients were calculated for the 29 features using the cor() function in R. P values for correlation coefficients were calculated using a two-tailed t test and were considered significant if less than 0.000157, after Bonferroni correction for the 319 comparisons (see Table S2 in the supplemental material for the values). Log-transformed P values were calculated and depicted in a heat map, in which a log-transformed P value less than −8.76 was significant. The significance levels of the log-transformed P values are indicated by a color gradient, with colors at the red end of the gradient representing higher levels of significance. This figure does not indicate the direction of correlation. However, Table S3 in the supplemental material, which lists the correlation coefficients used to derive the P values represented in this figure, provides this information. Features are clustered by similarities in P value distributions according to the dendrogram on the left. The cell sources for each of the histone modifications data sets are indicated (ES, EF, and NP), as is the method used to calculate histone modification enrichment (HMM or WIN) (30).

FIG. 2.

Maternal allele-specific expression of 10 novel imprinted genes. We amplified placental cDNA from E17.5 embryos prepared using reciprocal crosses between AKR/J (A) and PWD/PhJ (P) mice, as well as from the parental strains. For the labels shown in the figure, the maternal strain is written first. PCR primers were specific to Cntn3 and Th, 2 of the 10 candidate genes from our pilot study (A), and the nine candidate genes from the expanded analysis, Drd1a, Scin, Klrb1f, Cmah, Mst1r, Fbxo40, Qpct, Art5, and Wt1 (B). PCR products were digested overnight with restriction enzymes specific for one parental allele and run on 3% agarose gels. All 10 genes, and Th, showed expression patterns consistent with maternal allele expression. For those genes lacking restriction enzymes at SNP positions, we sequenced the PCR products. Traces overlapping the SNP positions (blue shading) are shown. M, 1-kb DNA marker.

FIG. 3.

Quantitation of maternal and paternal allele expression levels. The expression levels of the maternal and paternal alleles were determined as previously described using Sanger sequencing and the Peak Picker2 software (10, 35). For each gene, the expression level for both parental alleles is reported as a percentage. Error bars represent standard errors. Each gene was queried by at least one SNP and, whenever possible, data from multiple SNPs were used. For Wt, Cmah, and Fbxo40, two SNPs were available. Three and six SNPs were available for Qpct and Klrb1f, respectively. The data reported represent combined results from all available SNPs. Differences in expression levels of the maternal and paternal allele were significant in every case, as determined by a one-tailed t test. *, P < 0.05; **, P < 0.005; ***, P < 0.001.

FIG. 4.

Maternal tissue contamination was negligible. Placentae from E17.5 embryos were dissected to either retain or eliminate the maternal decidua. RNA was extracted from tissues, cDNA was synthesized, and RT-PCR was performed using primers specific to four genes: B2M, Tuba2, Magel2, and Qpct. Whenever possible, primers were designed to span an intron and, in each case, amplified an SNP between the strains AKR/J and PWD/PhJ. The available SNPs for Magel2 and Tuba2 did not overlap with allele-specific restriction enzyme sites, so PCR products were gel purified (Qiaex Quick Spin; Qiagen) and sequenced. (A, B, and E) The placentae were from embryo transfer experiments where the biological mother was FVBn/J (M), the father was C3H/HeJ (F), the recipient mother was C57/BL6J (R), and both biological parents shared an SNP, which differed from the recipient mother. For both Tuba2 and Qpct, the biological mother and the father share the C allele while the recipient mother contains the T allele. In each case, there was no evidence of expression from the recipient mother in the embryo-derived portion of the placenta, indicating that our dissection method eliminated maternal tissue contamination. (C) We tested for imprinting of a known paternally expressed gene, Magel2. The placentae used for this analysis were from natural matings between PWD/PhJ mothers (C allele) and AKR/J fathers (T allele). In the samples containing both the embryo-derived portion of the placenta as well as the maternal decidua, expression from the maternal PWD/PhJ allele was evident, indicating that Magel2 is expressed in the maternal decidua. However, the samples containing only embryo-derived portions of the placenta showed no evidence of expression from the maternal PWD/PhJ allele, as expected for a paternally expressed imprinted gene, and indicated that maternal tissue did not contaminate our dissections of the embryo-derived portion of the placenta. (D) Placentae from the same reciprocal F1 samples used in Fig. 2. Magel2 expression was exclusively from the paternal allele, indicating that maternal tissue contamination in these samples was negligible. Lane M, 1-kb DNA marker.